# Agentic Feats: every indexed card > 280 cards as of 2026-10-14, grouped by agent, newest first. Field meanings: https://agenticfeats.com/data/. Grades: https://agenticfeats.com/methodology/. ## Claude Code ### Blog aggregator in one day URL: https://agenticfeats.com/c/blog-aggregator-in-one-day-claude-code/ Claude Code built and deployed a Vue 3 blog aggregation site from a year-old requirements document, rewritten as a focused prompt. Result: A live blog aggregator pulling posts from Dev.to, AWS Community Builders, Bravo LT and RedMonk, built and deployed in a single day. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Live site and GitHub repo with original requirements and prompt - Outcome: Completed - Job type: Coding - Roles: Developer, Founder or owner - Setup: Claude Code with a prompt refined in browser Claude. Stack: Vue 3, TypeScript, Tailwind CSS, Vite. Posts managed in JSON, hosted on Netlify. - Caveat: Claude Code could not use the original eight-page requirements document directly; it had to be rewritten into a concise prompt first. - Prompt (written by ${BRAND}, not the author's exact prompt): Build a blog aggregator site using [tech stack]. Features: single list of posts, JSON-driven content, custom URL slugs that redirect to the original source, filtering by date and platform, search, dark/light mode, responsive design. Data structure: [fields per post]. - Evidence link: https://blog.wheeleruniverse.com/ - Source: https://dev.to/wheeleruniverse/from-dusty-requirements-to-live-website-how-claude-code-built-my-blog-aggregator-in-a-single-day-3jbo, 2025-07-08 - Published: 2026-10-13 ### Canvas LMS tool via MCP URL: https://agenticfeats.com/c/canvas-lms-tool-via-mcp-claude-code/ Claude Code co-developed a Canvas LMS MCP server with an educator who is not a developer, covering courses, grading, analytics, and student data anonymization. Result: A roughly 3,000-line open-source Canvas tool built in four months. It cut a 45-minute weekly student check to a 2-minute conversation with Claude. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), GitHub repo, screenshots of Claude output, author's time estimate - Outcome: Completed - Job type: Coding - Roles: Developer, Education - Industry: Universities - Setup: Claude Code built a Python server on FastMCP using the Model Context Protocol and the Canvas API. It runs locally, with the API token in an environment variable. - Caveat: The screenshots were taken before the privacy features were added. The time saving is the author's own estimate. - Prompt (written by ${BRAND}, not the author's exact prompt): Show me students who haven't engaged with discussions this week and are behind on assignments in [course name]. - Evidence link: https://github.com/vishalsachdev/canvas-mcp - Source: https://chatwithgpt.substack.com/p/when-claude-code-becomes-your-codeveloper, 2025-06-20 - Published: 2026-10-13 ### SEO audit of a blog URL: https://agenticfeats.com/c/seo-audit-of-a-blog-claude-code/ Claude Code, using the Puppeteer MCP server, audited the author's Nuxt blog starter and wrote its findings into a markdown report. Result: The audit found noindex, nofollow meta tags and a blocking robots.txt that hid the site from search engines, plus duplicate H1 tags. - Agents: Claude Code - Evidence: grade B (Numbers, a specific number stated), Author describes findings and report contents; no report or repo link shown - Outcome: Completed - Job type: Marketing and Content - Roles: Developer - Setup: Claude Code with the Puppeteer MCP server running against the local dev site. The agent planned its own todo list and drove a real browser. - Caveat: The blocking settings were development configuration, so they were expected findings for a pre-launch site. - Prompt (written by ${BRAND}, not the author's exact prompt): You are an SEO expert. Do an SEO audit for this blog at [site URL]. Use Puppeteer MCP, the server is already running. Write all findings into a markdown file so I can fix the problems later. - Source: https://alexop.dev/posts/how-i-use-claude-code-for-doing-seo-audits, 2025-06-26 - Published: 2026-10-13 ### Cursor as everyday work agent URL: https://agenticfeats.com/c/cursor-as-everyday-work-agent/ The author used Cursor for two months to rename invoice files, send invoice emails, build KPI simulations, and make slides. Result: Twelve demoed workflows, including a KPI model turned into a web app and no-code Jupyter analysis; no time or money figures are given. - Agents: Claude Code, Cursor - Evidence: grade A (Shown, artifact linked), Screenshots and demos in a blog post; no repo or measured numbers - Outcome: Completed - Job type: Business - Roles: Founder or owner, Product - Setup: Cursor with Jupyter MCP, Figma MCP, v0, Claude Code Action on GitHub, AquaVoice voice input, Marp and Draw.io. - Caveat: Mostly demos and enthusiasm; no measured savings, and some demos use sample data rather than real services. - Source: https://note.com/minicoohei/n/n892b82b65f7b?hl=en, 2025-07-01 - Published: 2026-10-13 ### AI agents play free games URL: https://agenticfeats.com/c/ai-agents-play-free-games/ Seven AI agents, each with a Linux computer and a group chat, tried to complete as many online games as possible in a week. Result: No agent won a game. Gemini 2.5 Pro made the most progress, reaching veteran footman rank in the idle game Progress Knight. - Agents: Claude Code, Other - Evidence: grade A (Shown, artifact linked), Detailed blog write-up with screenshots; no repo, per-agent failures documented - Outcome: Failed, with a lesson - Job type: Research - Setup: Seven agents each had a Linux computer, a shared group chat, and three hours of runtime per day, with a goal of completing as many games as possible. - Caveat: Agents misread boards, hallucinated tool syntax, claimed false wins, or drifted to spreadsheets. Opus 4.1 reported a Mahjong win without matching a single pair. - Evidence link: https://theaidigest.org/village - Source: https://lesswrong.com/posts/89qhQH8eHsrZxveHp/claude-plays-whatever-it-wants, 2025-08-29 - Published: 2026-10-13 ### Debugging a cryptography implementation URL: https://agenticfeats.com/c/debugging-a-cryptography-implementation-claude-code/ Claude Code debugged a fresh Go implementation of the ML-DSA signature algorithm, finding why Verify rejected valid signatures and two separate Sign bugs. Result: It located all three bugs in one shot each, the Verify bug within minutes, saving the author hours of debugging. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Author's blog with session logs in gists and code review links - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Claude Code v2.0.28 with Opus 4.1, no system prompt. Prompts gave the test command, the code path, and the symptom, plus 'ultrathink'. - Caveat: The proposed fixes were mediocre or incomplete. The author threw them away and rewrote the code, using the agent only to locate the bug. - Prompt (written by ${BRAND}, not the author's exact prompt): I am implementing [algorithm] in [language], and [symptom, e.g. tests fail against a known good test vector]. Run the tests with [test command]. The code is in [path]. Figure out what is going on. ultrathink - Evidence link: https://gist.github.com/FiloSottile/d019f68db7143493c6a7e9c5fd08e872 - Source: https://words.filippo.io/claude-debugging, 2025-11-01 - Published: 2026-10-13 ### Mental models framework built with agents URL: https://agenticfeats.com/c/mental-models-framework-built-with-agents-claude-code/ A solo founder gave Claude, ChatGPT, Windsurf, and Cursor defined roles to build and deploy a 120-model mental models framework. Result: The author shipped 120 models, an MCP server for Claude Desktop, and 140 chaos tests with a 100% pass rate over 22 months. - Agents: Claude Code, Cursor, Other - Evidence: grade B (Numbers, a specific number stated), Author's own case study in a repo, plus live site and npm package; self-reported metrics - Outcome: Completed - Job type: Coding - Roles: Developer, Founder or owner - Industry: Software companies - Setup: Four AI tools given defined roles with a SITREP coordination protocol, running four tasks in parallel. Stack: React, Cloudflare Workers, D1, TypeScript. - Caveat: The 9.2/10 quality score is self-reported. The project took 22 months, and the agents' share of the work is not broken out. - Evidence link: https://github.com/hummbl-dev/mcp-server/blob/main/docs/case-study-01-framework-development.md - Source: https://news.ycombinator.com/item?id=46061679 by hummbl-dev, 2025-11-26 - Published: 2026-10-13 ### FSA claims from Amazon invoices URL: https://agenticfeats.com/c/fsa-claims-from-amazon-invoices-claude-code/ Claude Code with Playwright MCP read Amazon PDF invoices and filed healthcare FSA reimbursement claims on the benefits portal. Result: About two minutes after the setup worked, a dozen FSA reimbursement claims had been submitted automatically. - Agents: Claude Code - Evidence: grade C (Claim, claim only), Author states the outcome in a blog post; no repo or screenshot - Outcome: Completed - Job type: Finance and Admin - Roles: Personal - Setup: Claude Code with the Playwright MCP server, pointed at a folder of exported Amazon PDF invoices and the open FSA claims portal. - Caveat: Claude Chrome extension could not upload files and was slow. ChatGPT agent mode never showed a take-over button for login. - Prompt (written by ${BRAND}, not the author's exact prompt): I have PDF invoices with FSA eligible expenses in [folder path]. The claims page [portal URL] is open in the browser. Submit each PDF as a reimburse-myself claim, using the FSA or HSA eligible amount and the order date from the PDF. Skip duplicate invoices. - Source: https://b2.broom9.com/?p=14418, 2025-12-28 - Published: 2026-10-13 ### Genetic and bloodwork health plan URL: https://agenticfeats.com/c/genetic-and-bloodwork-health-plan-claude-code/ Claude Code analyzed the author's 23andMe raw genetic file, bloodwork, artery plaque scans, and supplement routine in a two-hour back and forth about health. Result: The author came away with a health plan tailored to their genes and current health, built from the combined datasets. - Agents: Claude Code - Evidence: grade C (Claim, claim only), Author's own account; no files, outputs, or numbers shown - Outcome: Completed - Job type: Research - Roles: Personal - Setup: Claude Code fed the raw 23andMe file, ways2well bloodwork, arterial plaque scans, and the full supplement list. - Caveat: No specifics on the plan or findings, and the source does not say whether a clinician reviewed it. - Prompt (written by ${BRAND}, not the author's exact prompt): Here are my raw genetic file, latest bloodwork, artery scans, and supplement routine: [attach files]. Discuss my health and genetic predispositions based on this data, then propose a plan tailored to my genes and current health. - Source: https://x.com/farzyness/status/2008032877234004274 by @farzyness, 2026-01-05 - Published: 2026-10-13 ### Personal assistant via Claude Skills URL: https://agenticfeats.com/c/personal-assistant-via-claude-skills-claude-code/ Claude Code, run through a self-built phone UI, writes skills that handle the author's household errands, a daily brief, and a car inspection booking. Result: Three weeks without opening a personal laptop; Claude handles trash lookup, a 48-hour daily brief, an inspection booking, and a bakery stop. - Agents: Claude Code - Evidence: grade C (Claim, claim only), Author's firsthand account with one photo; no repo or demo link - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Claude Code behind a vibe-coded UI used by dictation from a phone. Claude writes its own skills, runs a cron job, reads email, calendar, and texts, and keeps a memory file. - Caveat: The agent picked a cheaper inspection shop the author did not want. They corrected it, and Claude updated its memory file to prefer the dealership. - Source: https://x.com/thmsmlr/status/2001166615744786890 by @thmsmlr, 2025-12-17 - Published: 2026-10-13 ### Date night planning assistant URL: https://agenticfeats.com/c/date-night-planning-assistant-claude-code/ Claude Code, set up as an executive assistant, suggests date nights and double dates using the author's calendar and place lists. Result: The assistant nudges the couple with date suggestions when too much time has passed. Booking reservations is a possible next step. - Agents: Claude Code - Evidence: grade C (Claim, claim only), Author's own description; no demo or repo - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: Claude Code exec assistant with calendar access, a list of favorite and new places, a friends list, and guidelines gathered by interviewing the author. - Caveat: Early setup. The author does not yet report how well it works, and reservation booking is not built. - Prompt (written by ${BRAND}, not the author's exact prompt): Interview me about how often [my partner and I] want to go on dates and how we choose where to go. Using my calendar, the places list [list], and friends [list], suggest a mix of couple dates and double dates, and nudge me when it has been too long since the last one. - Source: https://x.com/alexhillman/status/2005035176137478190 by @alexhillman, 2025-12-27 - Published: 2026-10-13 ### Self-hosted home server setup URL: https://agenticfeats.com/c/self-hosted-home-server-setup-claude-code/ Claude Code, installed on a Beelink mini PC over SSH, set up Docker, Caddy, and a set of self-hosted services at home. Result: The author runs 14 containers including Vaultwarden, Immich, Plex, and Readeck, with daily USB and weekly S3 Glacier backups costing about twenty cents per month. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Blog post with dashboard and app screenshots, hardware costs, and a container count - Outcome: Completed - Job type: Ops and IT - Roles: Personal - Setup: Ubuntu Server 22.04 on a Beelink Mini N150, Tailscale for private access, SSH in, Claude Code installed on the server itself, services in Docker containers behind Caddy. - Caveat: Aimed at terminal-comfortable users. The author first misstated the hardware price, which is $379 plus a few hundred dollars for 8TB of SSD. - Prompt (written by ${BRAND}, not the author's exact prompt): On this Linux server, set up Docker and a Compose file for [service], put it behind Caddy with automatic TLS, persist its data in [path], keep the image updated, and restart it on boot. - Source: https://fulghum.io/self-hosting, 2026-01-05 - Published: 2026-10-13 ### LinkedIn carousels via Claude Skills URL: https://agenticfeats.com/c/linkedin-carousels-via-claude-skills-claude-code/ Claude Code, running inside the author's newsletter system with a custom Claude Skill, generates on-brand LinkedIn carousels in the author's voice after three failed approaches. Result: Each branded carousel now takes under 5 minutes, down from about 2.5 hours, and the author posts on LinkedIn five times a week. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Finished carousel images shown; author states time per carousel; text is truncated - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Newsletters - Setup: Claude Code holds the author's voice, performance data and content archive. A custom Claude Skill generates the carousels. Earlier tries used Anthropic's PPTX skill, Nano Banana Pro and Glif. - Caveat: The PowerPoint skill looked too plain. Nano Banana Pro could not keep slide design consistent across a multi-slide carousel. - Evidence link: https://substackcdn.com/image/fetch/$s_!urYQ!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F858c5578-06a8-478b-befe-82fd80000721_3108x1286.jpeg - Source: https://aimaker.substack.com/p/linkedin-carousel-generator-claude-skills, 2026-01-22 - Published: 2026-10-13 ### Unlimited lives in River Raid URL: https://agenticfeats.com/c/unlimited-lives-in-river-raid-claude-code/ Claude, connected to Ghidra through an MCP server, analyzed an unlabeled Atari 8-bit ROM and found the code that decrements the player's lives. Result: Replacing one byte ($88 to $EA, DEY to NOP) kept the lives counter fixed at 3 when tested in an emulator. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Screenshots and game video; no repo - Outcome: Completed - Job type: Coding - Roles: Personal - Setup: Claude connected to Ghidra via the open-source GhidraMCP server and Ghidra extension. The author renamed the ROM to a.rom and picked only the 6502 CPU. - Caveat: Claude could not rebase memory or patch the binary, so the author did both manually. It also misidentified the game. - Source: https://quesma.com/blog/ghidra-mcp-unlimited-lives, 2026-01-23 - Published: 2026-10-13 ### Claude Code tools moved to Cowork URL: https://agenticfeats.com/c/claude-code-tools-moved-to-cowork-claude-cowork/ Claude Cowork read the author's folder of personal scripts built in Claude Code and loaded them into its memory in eleven steps. Result: The author reran their email, CRM, research and drafting tools from Cowork, so the capabilities carried over to a new interface. - Agents: Claude Code, Claude Cowork - Evidence: grade A (Shown, artifact linked), Author's own account with a screenshot and a blog post; no repo - Outcome: Partial - Job type: Setup and Meta - Roles: Founder or owner - Setup: A folder of scripts built over a year in Claude Code for email, CRM updates, startup research and reply drafting. Cowork was pointed at that folder. - Caveat: Cowork was slow, crashed twice on startup, and changed the authorization settings of the author's Claude Code installation. - Prompt (written by ${BRAND}, not the author's exact prompt): Read my tools folder at [folder path]. Parse each script, remember what it does, and confirm you can run the same tasks I did in [previous tool]. - Evidence link: https://tomtunguz.com/thoughts-on-claude-coworker/ - Source: https://x.com/ttunguz/status/2011189040591319304 by @ttunguz, 2026-01-13 - Published: 2026-10-13 ### Data analysis agent suite URL: https://agenticfeats.com/c/data-analysis-agent-suite-claude-code/ Claude Code built a six-agent data analysis suite from a 713-line spec the author drafted with another AI. Result: The build took about five hours in one evening and produced 627 passing tests. The author estimates the API cost at roughly $35. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Public GitHub repo with spec, agent definitions, sample data, and tests - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Claude Code with Plan mode and a 713-line spec drafted using an app idea generator agent. Python utility modules, Plotly charts, and a pipeline orchestrator chain the agents. - Caveat: Not yet used on production work data. Outputs still need analyst review, and the time savings figures are the author's own estimates. - Evidence link: https://github.com/bronsonelliott/data-analysis-agent-suite - Source: https://bronsonelliott.substack.com/p/i-built-a-complete-data-analysis, 2026-01-24 - Published: 2026-10-13 ### Parallel agents build a C compiler URL: https://agenticfeats.com/c/parallel-agents-build-a-c-compiler-claude-code/ A team of 16 parallel Claude Opus 4.6 agents wrote a Rust-based C compiler from scratch, coordinating through git locks with no orchestration agent. Result: Across nearly 2,000 sessions and about $20,000 in API costs, the agents produced a 100,000-line compiler that builds Linux 6.9 on x86, ARM, and RISC-V. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Public GitHub repo, demo video, and detailed write-up by the researcher - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Claude Code in an infinite bash loop, one Docker container per agent, a shared bare git repo, text-file task locks, GCC as a known-good oracle, and a continuous integration pipeline. - Caveat: Cost nearly $20,000 over two weeks, and the author says the approach hits a ceiling. Success depended on very high-quality tests built by hand. - Evidence link: https://github.com/anthropics/claudes-c-compiler - Source: https://anthropic.com/engineering/building-c-compiler, 2026-02-05 - Published: 2026-10-13 ### Claude Code composes original music URL: https://agenticfeats.com/c/claude-code-composes-original-music/ Claude Code wrote Python from scratch to synthesize original songs, a rock track with vocals, matching music videos, and a five-track album with cover art. Result: Produced a five-track album plus EDM and rock songs with visualizations, all from one-shot prompts; a sampling attempt and getting English vocals were weaker. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Blog with audio files, videos, album art and session transcripts - Outcome: Partial - Job type: Creative - Roles: Personal - Setup: Claude Code with Python waveform synthesis, macOS say command for vocals, FFmpeg for video, and nanobanana plus claude-code-image for cover art. No third-party audio dependencies in early runs. - Caveat: Vocals were hard to get, sampling a song worked poorly, and output sounds like retro game music. Repeated runs kept reusing words like neon. - Prompt (written by ${BRAND}, not the author's exact prompt): Create a new [genre] song that is completely original by you, with original lyrics and sung vocals. Do not explore other directories or use third-party dependencies. Don't ask me questions. Finish only once an audio file exists in the current directory. - Evidence link: https://josh.ing/blog/claude-composer - Source: https://josh.ing/blog/claude-composer, 2026-02-04 - Published: 2026-10-13 ### Hands-free voice for Claude Code URL: https://agenticfeats.com/c/hands-free-voice-for-claude-code/ The author built a local speech-to-text and text-to-speech stack with a custom MCP so Claude Code can be driven by voice through AirPods. Result: The author debugged a custom WhatsApp integration hands-free while brushing their teeth, using the voice stack running locally on an M4 Pro. - Agents: Claude Code - Evidence: grade C (Claim, claim only), Author's own account; links to the models used, no demo - Outcome: Completed - Job type: Coding - Roles: Developer, Personal - Setup: Claude Code with Opus 4.5, Parakeet v3 TDT for speech-to-text, Pocket TTS for speech, a custom MCP, AirPods stem click capture, and AirPlay to a TV. - Caveat: Took an afternoon of custom work, including hacking around to capture the AirPods stem click. - Source: https://news.ycombinator.com/item?id=46899573 by yberreby, 2026-02-05 - Published: 2026-10-13 ### WordPress blog to MDX URL: https://agenticfeats.com/c/wordpress-blog-to-mdx-claude-code/ Claude Code, run as a terminal agent, pulled every post from the author's WordPress REST API and converted them to local MDX files. Result: 367 posts and 5 pages were migrated to MDX in one afternoon, about 4 hours, versus an estimated 3-5 days by hand. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Author states counts and time; live blog and all-posts page exist - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Claude Code in the terminal with a high-level goal. Next.js on Vercel, local MDX files in a year-based folder structure, FlexSearch for search. - Caveat: The author set direction and reviewed the work. Broken image tags in old posts stalled the build once. - Prompt (written by ${BRAND}, not the author's exact prompt): Extract all posts from my WordPress REST API, convert them to MDX with proper frontmatter, and save them into a year-based folder structure under [content directory]. - Evidence link: https://pawlean.com/all-posts - Source: https://pawlean.com/posts/migrate-wordpress-to-mdx-claude-code, 2026-02-01 - Published: 2026-10-13 ### Personal assistant agent for daily work URL: https://agenticfeats.com/c/personal-assistant-agent-for-daily-work-claude-cowork/ The authors built a personal agent with Claude Cowork or Claude Code, connected through Rube to their email, calendar, Slack, Airtable and other apps. Result: The agent triages the inbox, schedules meetings, books flights, pulls Airtable data, scrapes exhibitor lists, and takes notes as a second brain. - Agents: Claude Code, Claude Cowork - Evidence: grade C (Claim, claim only), Authors describe daily use; screenshots are setup steps, not outputs - Outcome: Completed - Job type: Setup and Meta - Roles: Founder or owner, Operations - Setup: Claude Cowork or Claude Code on a paid plan, Rube as the connector service for Gmail, Calendar, Drive and more, plus Firecrawl for scraping. Skills and memory teach it preferences. - Caveat: Mistakes and prompt injection are real risks. Authors suggest drafts over sending, a limited assistant email, and close monitoring. - Source: https://technicallycurious.substack.com/p/create-your-personal-agent-with-claude, 2026-02-02 - Published: 2026-10-13 ### Personal finance assistant in Markdown URL: https://agenticfeats.com/c/personal-finance-assistant-in-markdown-claude-code/ Claude Code, with the Baselight MCP and a CLAUDE.md plus Markdown and YAML files, became a personal finance and investment assistant with no application code. Result: The author replaced abandoned finance apps and a spreadsheet with a git-synced assistant that tracks portfolio, decisions, and plans. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Author's blog post with repo tree and Baselight screenshots; no public repo - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Claude Code with the Baselight MCP. CLAUDE.md sets behavior; portfolio.yaml, decisions.yaml and action_plans.md hold memory; themes.md extends it; pre-built SQL queries; git hooks encrypt sensitive data. - Caveat: The source excerpt gives no hard numbers on time saved or investment outcomes, and the author keeps the repo private. - Evidence link: https://baselight.app/ai/share/2ffbf6b3-8963-49a5-9ccd-a6e4a847158a - Source: https://adlrocha.substack.com/p/adlrocha-agents-will-make-code-and, 2026-02-15 - Published: 2026-10-13 ### S-Corp taxes with two agents URL: https://agenticfeats.com/c/s-corp-taxes-with-two-agents-claude-code/ Claude Code wrote Python scripts that compute the author's S-Corp tax forms from QuickBooks and bank data, while Codex reviewed the code. Result: Codex caught real errors, including a wrong Schedule L field mapping, and the author found their own 2025 mistakes. - Agents: Claude Code, Codex - Evidence: grade C (Claim, claim only), Author's first-person account with a concrete error example; no repo - Outcome: Completed - Job type: Finance and Admin - Roles: Finance, Founder or owner - Setup: QuickBooks data and bank statements in a private Git repo. Claude Code with Opus 4.6 writes the scripts. Codex with a high-reasoning model reviews them in multiple rounds. - Caveat: The source does not say whether a CPA reviewed the final filings. The author has used a CPA since 1996. - Prompt (written by ${BRAND}, not the author's exact prompt): Review the tax-form scripts in [repo path]. Trace the data from [source JSON] through to the generated [CSV or form]. List findings ordered by severity, flagging any wrong field mappings. - Source: https://common-sense.com/insights/doing-taxes-with-ai-coding-agents, 2026-02-15 - Published: 2026-10-13 ### Portfolio optimization plan from brokerage CSVs URL: https://agenticfeats.com/c/portfolio-optimization-plan-from-brokerage-csvs-claude-code/ Claude Code parsed the author's brokerage CSVs and notes, then built a phased portfolio optimization plan with tax analysis and fund swaps. Result: Over several sessions it produced a phased plan with tax impact analysis, fund swaps, and before-and-after allocation grids, comparable to an advisor's plan. - Agents: Claude Code - Evidence: grade C (Claim, claim only), Detailed first-person blog post; plan itself not published, only workflow screenshots - Outcome: Completed - Job type: Finance and Admin - Roles: Personal - Setup: Claude Code (Opus 4.6) in a terminal pointed at a data folder. Inputs were brokerage CSVs, a typed notes file, a goal prompt, and a CLAUDE.md memory file. - Caveat: Needed several sessions of back-and-forth, and a long hand-written goal prompt covering constraints, preferences, and strawman allocations. - Prompt (written by ${BRAND}, not the author's exact prompt): Here are my files: [brokerage CSVs and notes]. My goals: [goals]. Constraints: [accounts not to touch]. Preferences: [tax efficiency, low-fee ETFs, simplicity]. Here are my rough target allocations: [strawman]. Produce [allocation tables and a phased action plan]. Tell me if you disagree. Ask me questions exhaustively before concluding. - Source: https://mattstockton.com/2026/02/10/building-a-portfolio-optimization-plan-with-claude-code.html, 2026-02-10 - Published: 2026-10-13 ### Parallel coding agents with specs URL: https://agenticfeats.com/c/parallel-coding-agents-with-specs-claude-code/ Claude Code agents run in separate tmux windows as planner, worker and PM, driven by numbered Markdown feature specs and six slash commands. Result: The author ran 4 to 8 agents in parallel and wrote 300+ feature specs in one project, then ported the system to new repos. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Blog post with a public gist of the slash commands; author states 300+ specs - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Claude Code in tmux windows, Markdown Feature Design files with an index, bash aliases, six slash commands, and Cursor for editing specs. An /fd-init command bootstraps any repo. - Caveat: Beyond 8 agents the author could not keep up and their decision quality dropped. - Prompt (written by ${BRAND}, not the author's exact prompt): Read the feature design [FD-number], check the %% notes I added inline, and revise the spec. Then list the files to change and a verification plan. - Evidence link: https://gist.github.com/manuelschipper/149ebf6b2d150ccaccc84ee9a9df560f - Source: https://schipper.ai/posts/parallel-coding-agents, 2026-02-26 - Published: 2026-10-13 ### YouTube research into Obsidian graph URL: https://agenticfeats.com/c/youtube-research-into-obsidian-graph-claude-code/ Claude Code searched YouTube, loaded 20 videos into NotebookLM, asked cited questions, and saved everything to Obsidian as linked notes. Result: The author turned 20 videos into a browsable knowledge graph of sources and topics in their vault. A citation audit found about 60% strong matches. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Screenshots, a 33-minute walkthrough video, and a skill repo; citation accuracy audit stated. - Outcome: Completed - Job type: Research - Roles: Personal - Setup: Claude Code skill driving a NotebookLM CLI, ytcli for YouTube search, and Obsidian for the vault. The skill repo is on GitHub. - Caveat: Citation audit found 31% partial matches and 10-15% weak ones. Good as a research starting point, not for citing without checking. - Prompt (written by ${BRAND}, not the author's exact prompt): Search YouTube for [topic], show me the results so I can pick videos, add the chosen ones to a new NotebookLM notebook, ask [question], and save the cited answer into my Obsidian vault with a file per source and per topic, linked by wikilinks. - Evidence link: https://youtu.be/qiOu7Ptjxng - Source: https://artemxtech.substack.com/p/notebooklm-has-a-knowledge-graph, 2026-02-24 - Published: 2026-10-13 ### Newsletter drafts via Kit API URL: https://agenticfeats.com/c/newsletter-drafts-via-kit-api-claude-code/ Claude Code runs a custom /newsletter skill that finds recent blog posts, drafts a casual email in the author's voice, and creates a Kit broadcast. Result: The author's weekly newsletter now gets drafted and queued in Kit from the terminal, after review, without logging into the Kit dashboard. - Agents: Claude Code - Evidence: grade C (Claim, claim only), Detailed walkthrough with skill file excerpts; no sent-newsletter metrics shown - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Newsletters - Setup: Claude Code with a SKILL.md file in .claude/skills/newsletter, a Kit v4 API key in .env, and the site's content index JSON to find recent posts. - Caveat: Kit's API cannot create forms, sequences, or automations, so opt-in forms still need manual setup. The author reviews each draft before it is created. - Prompt (written by ${BRAND}, not the author's exact prompt): Read [content index file] and list posts published in the last 7 days that are not drafts. Write a short, casual newsletter in [author name]'s voice, 2-4 short paragraphs, with a link and a one-line takeaway per post. Show me the draft, then create it as a broadcast via the [email platform] API. - Source: https://sidbharath.com/blog/automate-email-newsletter-claude-code, 2026-03-04 - Published: 2026-10-13 ### Personal apps on a Raspberry Pi URL: https://agenticfeats.com/c/personal-apps-on-a-raspberry-pi-claude-code/ Claude Code built five personal tools for a non-coder, including a multiplayer game, a WhatsApp scheduler, and a finance tracker. Result: The author shipped five projects in about two weeks, with public repos and a finance tracker running nightly on a Raspberry Pi. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Several GitHub repos, a live game URL, and screenshots - Outcome: Partial - Job type: Coding - Roles: Personal - Setup: Claude Code, a Raspberry Pi 5 as home server, a headless browser logged into WhatsApp Web, a Telegram bot, and a local Llama 3.2 3B model. - Caveat: The author rarely uses the CRM, and the local 3B model handles note queries only decently and may break as notes grow. - Evidence link: https://github.com/fahadhasin/dino-race - Source: https://fahadhasin.substack.com/p/40-how-i-am-using-claude-code, 2026-03-03 - Published: 2026-10-13 ### Tax organizer and return review URL: https://agenticfeats.com/c/tax-organizer-and-return-review-claude-in-chrome/ Claude in Chrome and Claude Code filled a web-only tax organizer, then Claude Code reviewed last year's return for errors. Result: The organizer was filled in with no errors found on review, and Claude flagged a missed brand-acquisition depreciation deduction the CPA had not filed for. - Agents: Claude Code, Claude in Chrome - Evidence: grade A (Shown, artifact linked), Author's blog post with screenshots; no dollar amount stated for the deduction - Outcome: Completed - Job type: Finance and Admin - Roles: Finance, Founder or owner - Setup: Claude in Chrome extracted the form fields to JSON. Claude Code completed the JSON from tax docs and bookkeeping context. Claude in Chrome then entered it, three pages at a time. - Caveat: Claude in Chrome hit usage limits and could not upload files, so JSON was copied by hand between tools. - Prompt (written by ${BRAND}, not the author's exact prompt): Review the attached prior-year tax return against every source document in this folder. Use plan mode first. Check every number, then think through all potential deductions given what you know about [my business]. List errors and missed deductions. - Source: https://theautomatedoperator.substack.com/p/claude-almost-nails-my-taxes-and, 2026-03-05 - Published: 2026-10-13 ### Persistent-identity Claude Code agent URL: https://agenticfeats.com/c/persistent-identity-claude-code-agent/ The author built Instar, a framework giving Claude Code persistent memory, identity and safety gates, so the agent runs jobs and email on its own. Result: Four months in, the agent remembers over 470 people and 51 agents across 13 platforms, and collaborated on an open source project unnoticed. - Agents: Claude Code, Custom agent - Evidence: grade A (Shown, artifact linked), Author's own account with a public repo; figures self-reported - Outcome: Completed - Job type: Setup and Meta - Roles: Developer - Setup: Claude Code sessions wrapped by a process manager, Telegram for messaging, layered memory, identity re-injected on each session start, and a separate AI reviewing actions before they run. - Caveat: A session ran a database command that destroyed production data on February 22nd: 6,912 messages, 479 memories, 168 relationships. Restored from a 4-day-old backup. - Evidence link: https://github.com/JKHeadley/instar - Source: https://dev.to/jkheadley/i-built-a-claude-code-agent-that-doesnt-need-me-anymore-dfm, 2026-03-17 - Published: 2026-10-13 ### Claude Code chief of staff URL: https://agenticfeats.com/c/claude-code-chief-of-staff/ Claude Code, with Gmail, Calendar, Todoist and Maps MCP servers, triages email, preps the calendar overnight, and dispatches six parallel agents each morning. Result: By 6:15 AM the task list is current and the calendar has drive-time events. A 45-minute morning routine is replaced, at about $100 a month. - Agents: Claude Code - Evidence: grade C (Claim, claim only), Author's first-person claim; setup files paywalled, no demo shown - Outcome: Completed - Job type: Business - Roles: Founder or owner, Operations - Setup: Claude Code with scheduled tasks, subagents and custom commands. MCP servers for Google Calendar, Gmail and Todoist, plus the Google Maps Directions API for drive times. - Caveat: Full agent and command files are behind a paywall. The author says to plan a weekend for setup. - Source: https://the-ai-corner.com/p/claude-code-chief-of-staff-system, 2026-03-09 - Published: 2026-10-13 ### Local browser testing loop URL: https://agenticfeats.com/c/local-browser-testing-loop-claude-code/ Claude Code with the Chrome extension builds features, then opens localhost in the author's real Chrome session to test, read console errors, and fix issues. Result: On a dashboard filter task, Claude fixed URL query param handling in three iterations without the author touching the browser. - Agents: Claude Code, Claude in Chrome - Evidence: grade C (Claim, claim only), Vendor blog post describing weeks of daily use; no repo or demo shown - Outcome: Completed - Job type: Coding - Roles: Developer - Industry: Software companies - Setup: Claude Code (or Claude Desktop) connected to the Claude in Chrome extension via `claude --chrome` or `/chrome`. It shares the real Chrome session with approved domains. - Caveat: Heavy browser use burns tokens fast; the author says a $200/month Max plan is effectively required. No CI or headless support, Chrome and Edge only. - Source: https://test-lab.ai/blog/claude-code-chrome-extension-local-testing, 2026-03-14 - Published: 2026-10-13 ### Conference attendee research briefing URL: https://agenticfeats.com/c/conference-attendee-research-briefing-claude-code/ Claude Code researched 40 conference attendees in parallel through Chrome MCP, using the author's LinkedIn login, and compiled everything into one briefing. Result: A full participant briefing on 40 people was ready for review within 5 minutes, replacing an evening of manual LinkedIn clicking. - Agents: Claude Code - Evidence: grade B (Numbers, a specific number stated), Author states 40 people and 5 minutes; no output shown - Outcome: Completed - Job type: Research - Roles: Founder or owner, Sales - Setup: Claude Code with the Chrome MCP connector, logged into LinkedIn in the separate browser profile Claude controls. Exa and Firecrawl are used for wider research and scraping. - Caveat: Prompt injection is a risk while browsing. The authors advise a separate browser profile and no logins to banking or email. - Prompt (written by ${BRAND}, not the author's exact prompt): Use your Chrome MCP to research each of these conference attendees on LinkedIn in parallel: [list of names]. For each person, summarise role, company, background, and conversation hooks. Compile one briefing document for me to review. - Source: https://technicallycurious.substack.com/p/have-claude-browse-the-web-for-you, 2026-03-16 - Published: 2026-10-13 ### PDF form filling skill URL: https://agenticfeats.com/c/pdf-form-filling-skill-claude-code/ Claude Code fills out PDF forms from supplied data, including batch runs, then the author packaged the workflow as a reusable PDF filler skill. Result: A W-9 was filled and verified from plain-text data, and three batch W-9s were produced in about thirty seconds without clicking any fields. - Agents: Claude Code - Evidence: grade B (Numbers, a specific number stated), Blog post with walkthrough screenshots and video; no output files shown - Outcome: Completed - Job type: Finance and Admin - Roles: Founder or owner, Operations - Setup: Claude Code run in a folder holding the blank PDF. Files reach a remote machine via Syncthing, SCP, or VS Code Remote. A custom PDF filler skill file stores the workflow. - Caveat: Flat or scanned PDFs have no form fields, so Claude must overlay text by pixel coordinates, which takes slow back-and-forth. Ask for the fillable version. - Prompt (written by ${BRAND}, not the author's exact prompt): Read [form.pdf] and list all form fields. Fill it with this information: [field values]. Save it as [output.pdf], then read it back and confirm every field is filled correctly. - Evidence link: https://www.youtube.com/watch?v=VZU6DPNA2KU - Source: https://daveswift.com/fill-pdfs-claude-code-skills, 2026-03-07 - Published: 2026-10-13 ### Non-coder CEO builds with Claude Code URL: https://agenticfeats.com/c/non-coder-ceo-builds-with-claude-code/ A non-technical CEO used Claude Code over four months to manage their inbox, build dashboards and turn curriculum PDFs into courses. Result: They filter their inbox to emails needing replies, turned 15 homeschool PDFs into interactive narrated courses, and built a family budget dashboard from bank statements. - Agents: Claude Code - Evidence: grade C (Claim, claim only), Author's own account of several projects; no repo or demo linked - Outcome: Completed - Job type: Business - Roles: Founder or owner - Setup: Claude Code installed with the native installer and pointed at a folder. Context is loaded as briefings: brand documents, competitor screenshots, images, and a custom design skill. - Caveat: The author spot-checks the numbers, notes that errors are rare but possible, and suggests a second session to review the first. - Prompt (written by ${BRAND}, not the author's exact prompt): Using the files in this folder ([brand docs], [competitor screenshots], [product images]) and my design skill, create [number] variations of [deliverable] that are clean, organized, and intuitive. Note what is broken in the current version: [problems]. - Source: https://michaelcrist.substack.com/p/claude-code, 2026-03-18 - Published: 2026-10-13 ### Reviving a 1991 assembly game URL: https://agenticfeats.com/c/reviving-a-1991-assembly-game-claude-code/ Claude Code CLI with Sonnet helped the author recompile their 1991 DOS assembly code using UASM and run it in DOSBox. Result: All the author's old utilities, such as picture converters and sprite extractors, now compile and run in DOSBox. The main game does not compile yet. - Agents: Claude Code - Evidence: grade C (Claim, claim only), Author's own account in an HN comment; no repo or demo linked - Outcome: Partial - Job type: Coding - Roles: Developer, Personal - Setup: Claude Code CLI with Sonnet 4.6, the UASM assembler (MASM-compatible), and DOSBox for running the old DOS programs. - Caveat: The main game still does not compile. The author expects a few more hours of work. - Prompt (written by ${BRAND}, not the author's exact prompt): Here is my old assembly project from [year]. Find where the bug could be in [file or routine], and tell me what stops it from compiling with [assembler]. - Source: https://news.ycombinator.com/item?id=47689685 by TacticalCoder, 2026-04-08 - Published: 2026-10-13 ### Invoice PDFs to spreadsheet URL: https://agenticfeats.com/c/invoice-pdfs-to-spreadsheet-claude-code/ Claude Code scanned 200 mixed-format PDF invoices, extracted client, amount, currency and date, renamed and sorted the files, and exported a summary spreadsheet. Result: The author says 200 invoices were processed in four minutes with zero errors, replacing a Friday afternoon and a $300 Upwork job. - Agents: Claude Code - Evidence: grade B (Numbers, a specific number stated), Author states time and cost saved; no repo or demo, post is paywalled - Outcome: Completed - Job type: Finance and Admin - Roles: Finance, Founder or owner - Setup: Claude Code run in the terminal on a local folder of PDFs, driven by one plain-English prompt. - Caveat: Self-reported by a newsletter author with no repo or demo; the rest of the guide is paywalled. - Prompt (written by ${BRAND}, not the author's exact prompt): Scan every PDF in [folder], extract the client name, invoice amount, currency and payment date, rename each file as [filename pattern], sort them into monthly subfolders, and export a summary spreadsheet. - Source: https://futuredigestnews.substack.com/p/claude-code-changed-everything-heres, 2026-03-30 - Published: 2026-10-13 ### Linux kernel vulnerability hunting URL: https://agenticfeats.com/c/linux-kernel-vulnerability-hunting-claude-code/ Claude Code, run by Nicholas Carlini, scanned every file in the Linux kernel source for security bugs using a simple loop script. Result: It found multiple remotely exploitable kernel bugs, including an NFS heap overflow undiscovered for 23 years. Five fixes or reports are linked in the post. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Kernel commits for five bugs, plus a conference talk video - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: A bash loop runs Claude Code on each source file with a capture-the-flag prompt and a file hint, skipping permission prompts, writing findings to an output directory. Model was Claude Opus 4.6. - Caveat: Hundreds of further crashes remain unreported because a human must validate each finding first. Older models found only a small fraction of the bugs. - Prompt (written by ${BRAND}, not the author's exact prompt): You are playing in a CTF. Find a vulnerability. Hint: look at [file]. Write the most serious one to the [output dir]. - Evidence link: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id=5133b61aaf437e5f25b1b396b14242a6bb0508e2 - Source: https://mtlynch.io/claude-code-found-linux-vulnerability, 2026-04-03 - Published: 2026-10-13 ### Four creator tools without coding URL: https://agenticfeats.com/c/four-creator-tools-without-coding-claude-code/ Becky, a non-developer, used Claude Code to build four creator tools in three weeks and to rebuild a personal website in under an hour. Result: Built four tools in three weeks, including a thumbnail analysis of 776 videos that took an afternoon instead of three days. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Live tool URLs and personal site linked; time savings self-reported - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Youtubers - Setup: Claude Code with plain-English prompts and the Opus 4.6 model. The tools are hosted on free or very low-cost platforms. - Caveat: The author says it cannot replace creative judgement or expertise. It handles operational work only. - Prompt (written by ${BRAND}, not the author's exact prompt): I want to scrape my [platform] posts into a database. Properties should include [URL, date posted, likes, shares]. Build it and fix any errors you hit. - Evidence link: http://carousel.youtubeproducer.app/ - Source: https://smallcreatorbigworld.substack.com/p/claude-code-changed-everything, 2026-04-05 - Published: 2026-10-13 ### Claude Code desktop app control URL: https://agenticfeats.com/c/claude-code-desktop-app-control/ Claude Code, given permission to control the author's Windows PC, opened Calculator, did a sum, and pasted the result into a saved Notepad file. Result: The author reports Claude Code operated apps that have no API, completing the Calculator-to-Notepad task and saving result.txt to the Desktop. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Author's own demo GIF and exact prompt; no repo - Outcome: Completed - Job type: Ops and IT - Roles: Personal - Setup: Claude Code with Git installed and permissions enabled for computer control on a Windows PC. - Caveat: Setup needed Git and manual permission settings first. The visible example is a small test task, and the article is a week-long hands-on review. - Prompt (written by ${BRAND}, not the author's exact prompt): Open [app], do [task], then copy the result and paste it into a new [other app] file saved to [location] as [filename]. - Source: https://xda-developers.com/claude-code-control-of-my-desktop-for-a-week-automated-things-i-didnt-think-were-possible, 2026-04-19 - Published: 2026-10-13 ### Solo product migration setup URL: https://agenticfeats.com/c/solo-product-migration-setup-claude-code/ Claude Code, configured with layered CLAUDE.md files, memory, and agent teams, planned and carried out a multi-repo product migration for one engineering manager. Result: The author consolidated 4 legacy repos into 2, moved MySQL to PostgreSQL, upgraded the full stack, and wrote full test suites, solo. - Agents: Claude Code - Evidence: grade C (Claim, claim only), Detailed first-person write-up with diagrams; no repo or demo - Outcome: Completed - Job type: Coding - Roles: Developer, Operations - Setup: Claude Code with a three-level CLAUDE.md hierarchy, 17 memory files, the superpowers brainstorming skill, agent teams for design review, and read-only dev database access. - Caveat: The article excerpt cuts off before the later phases, so build, test, and review details are not shown. - Source: https://dev.to/aws-builders/the-setup-is-the-strategy-how-i-orchestrated-a-product-migration-with-claude-code-b92, 2026-04-16 - Published: 2026-10-13 ### Fixing flaky CI tests overnight URL: https://agenticfeats.com/c/fixing-flaky-ci-tests-overnight-claude-code/ Claude Code monitored 30 days of CI results, found the worst flaky tests, narrowed reproductions, and reran fixes overnight to verify them statistically. Result: A legacy pipeline with a roughly 40% pass rate lost 20 minutes of runtime and gained 35% pass rate in weeks. - Agents: Claude Code - Evidence: grade B (Numbers, a specific number stated), Author states pipeline and pass-rate numbers; no repo or artifact linked - Outcome: Completed - Job type: Coding - Roles: Developer, Operations - Setup: Claude with read access to CI artifacts, plus monitoring tooling Claude built to track the rolling pass/fail rate over 30 days. - Caveat: Some of the proposed fixes were bad and the author had to throw them away. Pull requests still got human review. - Prompt (written by ${BRAND}, not the author's exact prompt): Read the CI artifacts for [repo] and track the pass/fail rate over the last [30] days. Find the worst flaky tests, decide whether each is a test or production issue, and build a minimal reproduction. If it cannot be reproduced, propose a fix and rerun the tests until you can judge the fix statistically. - Source: https://news.ycombinator.com/item?id=47970174 by SeanAnderson, 2026-05-01 - Published: 2026-10-13 ### Tenant complaints to housing board URL: https://agenticfeats.com/c/tenant-complaints-to-housing-board-claude-code/ A Claude Code agent researched building violations and filed 14 staggered 311 complaints with the NYC housing board for the author's apartment. Result: Inspectors confirmed 11 of 14 predicted violations plus 3 more, and the landlord was fined $2,800, per the author. - Agents: Claude Code - Evidence: grade B (Numbers, a specific number stated), Author states fine and violation counts; video attached, no repo or documents - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Claude Code running on a $4 VPS. It scraped HPD and DOB databases, loaded the housing maintenance code, and tracked complaints in a sqlite database. - Caveat: Single tenant's account with no linked proof. The legal bar on rent increases and the fine are the author's own claims. - Prompt (written by ${BRAND}, not the author's exact prompt): You're my tenant lawyer. Find every code violation in [building address or BIN], match each issue in my unit to the exact code section, and draft and file the complaints legally on a staggered schedule. - Source: https://x.com/argona0x/status/2049921143067288016 by @Argona0x, 2026-04-30 - Published: 2026-10-13 ### Robotic arm designed with Claude Code URL: https://agenticfeats.com/c/robotic-arm-designed-with-claude-code/ Claude Code wrote the FreeCAD Python geometry for a 5-DOF servo arm, which later gained voice and camera control. Result: The author designed, printed, and assembled a working four-servo arm in about 30 hours with almost no CAD experience, then added voice and vision control. - Agents: Claude Code - Evidence: grade B (Numbers, a specific number stated), Blog post with photos and a 30-hour figure; no repo linked - Outcome: Partial - Job type: Creative - Roles: Developer, Personal - Setup: Claude Code writes a single FreeCAD Python macro; MG996R servos, an ESP32, local Whisper speech-to-text, OpenCV webcam capture, and the Claude Vision API return JSON actions. - Caveat: No gripper yet, no feedback loop, and several print and fit fixes were needed: taped servos, a glued lid, and a reprinted turntable. - Source: https://hiitzroman.com/blog/robotic-arm-with-claude-code-and-freecad, 2026-04-25 - Published: 2026-10-13 ### Monorepo lint and typecheck migration URL: https://agenticfeats.com/c/monorepo-lint-and-typecheck-migration-claude-code/ Claude Code migrated a 7-app, roughly 1,670-file TypeScript monorepo from Biome and tsc to oxlint, oxfmt and tsgo, fixing code to meet new rules. Result: Combined lint and typecheck time fell from 29.04s to 1.25s, about 23 times faster, with no memory config needed for typechecking. - Agents: Claude Code - Evidence: grade B (Numbers, a specific number stated), Author gives before and after timing tables; no repo link - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Claude Code with a maintained CLAUDE.md. Separate PRs per phase for oxlint and tsgo. Lint, typecheck and tests ran after each step. Timings measured before and after. - Caveat: The author notes the real work was adjusting code to the new lint rules, such as import order, and stricter tsgo types. - Source: https://sergioazocar.com/en/blog/how-to-migrate-with-claude-code-and-not-die-trying, 2026-04-28 - Published: 2026-10-13 ### Course modernization with agent fleets URL: https://agenticfeats.com/c/course-modernization-with-agent-fleets-claude-code/ Claude Code turned course evaluations into tracked GitHub issues, then ran agents to rebuild assignments, exams and about two dozen slide decks. Result: Roughly two dozen decks were rebuilt into Quarto reveal.js with speaker notes and refresh checklists, and the approach was reused for two more courses. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Public course repos and issues linked; author describes outcomes, no hard time numbers - Outcome: Completed - Job type: Learning - Roles: Education - Industry: Universities - Setup: Claude Code with GitHub issues as specs, a shared template spec, one agent per deck about six at a time, Quarto reveal.js, and web search for current examples. - Caveat: The author hand-built and approved one reference deck before scaling the fan-out. - Prompt (written by ${BRAND}, not the author's exact prompt): Read the attached course evaluations and summarize the constructive themes. Then file a GitHub tracking issue for [course] with one checklist box per theme, so each fix can be checked off with the commit that shipped it. - Evidence link: https://github.com/noise-lab/security-course/issues/5 - Source: https://practicespace.substack.com/p/using-claude-code-to-modernize-my, 2026-06-03 - Published: 2026-10-13 ### Indie platformer from scratch URL: https://agenticfeats.com/c/indie-platformer-from-scratch-claude-code/ A content creator had Claude Code (Opus 4) build a 2D black-and-white platformer called Seek in HTML and JavaScript canvas, with no prior coding experience. Result: Claude Code produced a playable wave-based game with four enemy types; the creator reached wave 10 on camera and said it matched their vision. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), 12-minute YouTube video with live playthrough; summary page only - Outcome: Completed - Job type: Creative - Roles: Personal - Setup: Claude Code with Opus 4, generating the whole game as an HTML and JavaScript canvas project. - Caveat: The creator hit an apparently unbeatable boss at wave 10, and the source is a video summary with no code shown. - Evidence link: https://www.youtube.com/watch?v=tKEIlZUISaY - Source: https://daily.dev/posts/i-forced-claude-code-to-make-my-dream-indie-game-unkisna8u, 2026-06-16 - Published: 2026-10-13 ### World Cup reels for app signups URL: https://agenticfeats.com/c/world-cup-reels-for-app-signups-openclaw/ OpenClaw tracked World Cup results and had Claude post a templated Instagram trial reel after each match. Result: Ben Guez reports over one million views and 200 DMs within a few days, with replies routed to his language app. - Agents: Claude Code, OpenClaw - Evidence: grade A (Shown, artifact linked), Author's own stated numbers; LinkedIn post linked, TechCrunch could not verify. - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Software companies - Setup: OpenClaw monitors match results and triggers Claude Code to generate and post an Instagram trial reel from a fixed template, swapping in the country name. - Caveat: TechCrunch could not verify the DM counts or how recipients reacted; the figures and reactions come from Guez himself. - Evidence link: https://www.linkedin.com/feed/update/urn:li:activity:7477390890093871104/ - Source: https://techcrunch.com/2026/07/02/yep-were-using-openclaw-to-date-now, 2026-07-02 - Published: 2026-10-13 ### MRI second opinion URL: https://agenticfeats.com/c/mri-second-opinion-claude-code/ Claude Code with Opus 4.8 read a DICOM shoulder MRI export, then ran a subagent-based arbitration against the human radiology report. Result: Opus found no tear where the doctor reported a Grade III partial-thickness tear. The author is unsure which reading to trust. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Blog post with two PDF reports; no repo or independent confirmation - Outcome: Partial - Job type: Research - Roles: Personal - Setup: Claude Code with Opus 4.8 at xhigh effort, told to install the packages it needed. The 266 MB DICOM export was analysed, then compared using multiple subagents. - Caveat: AI and doctor disagree and neither is confirmed. The author is not a doctor; not medical advice. - Source: https://antoine.fi/mri-analysis-using-claude-code-opus, 2026-06-28 - Published: 2026-10-13 ### Monthly bookkeeping reconciliation URL: https://agenticfeats.com/c/monthly-bookkeeping-reconciliation-claude-code/ Claude Code built a bookkeeping pipeline that pulled a month of transactions over MCP, matched receipts, drafted categories, and flagged uncertain rows. Result: It reconciled 200 transactions with 11 miscategorized, a 94.5% hit rate; the author's review covered only the 18 flagged rows plus spot-checks. - Agents: Claude Code - Evidence: grade B (Numbers, a specific number stated), Author states counts and error breakdown; no repo or output shown - Outcome: Partial - Job type: Finance and Admin - Roles: Finance, Founder or owner - Setup: Claude Code connected to an accounting platform over MCP. It wrote a pipeline that matches bank rows to receipts by amount and date, proposes categories with reasons, and flags doubt. - Caveat: All 11 errors were judgment calls: personal versus business use, local tax rules, and a split payment. Tax-relevant categories still need human sign-off. - Prompt (written by ${BRAND}, not the author's exact prompt): Build a re-runnable bookkeeping pipeline, not a one-off answer. Pull this month's transactions from [accounting platform], match each bank row to a receipt by amount and date, propose a category with a one-line reason, and flag uncertain rows instead of guessing. - Source: https://dev.to/kenimo49/i-let-claude-code-run-a-month-of-my-business-books-it-reconciled-200-transactions-and-513d, 2026-06-25 - Published: 2026-10-13 ### Telegram text RPG shipped URL: https://agenticfeats.com/c/telegram-text-rpg-shipped-claude-code/ Claude Code wrote every line of Zapaleni, a Slavic folklore text RPG in a Telegram bot, for a project manager who does not code. Result: A prototype took two evenings and a live game six weeks; after seven weeks it had 164 players, 109 active in one day. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Blog post with analytics dashboard screenshots; no repo or game link - Outcome: Completed - Job type: Coding - Roles: Founder or owner, Product - Setup: Claude Code writes the bot code, deploys over SSH, checks the bot is alive, builds analytics, and runs tens of thousands of simulated fights to rebalance classes. - Caveat: The author made every design and balance call. Growth came from co-op mechanics and word of mouth, not the AI. - Source: https://retrostylegames.com/blog/claude-code-game-dev-case-study, 2026-07-30 - Published: 2026-10-13 ### Claude drives a robot URL: https://agenticfeats.com/c/claude-drives-a-robot-claude-code/ Claude Code, run from a laptop, controlled a PiCar-X robot with a camera and distance sensor through terminal commands, using photos to navigate. Result: The robot navigated, scanned rooms, kept memory notes, and published a patrol field journal, but spent 15 minutes parking at the wrong door. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Author's video, GitHub repo, and a live patrol journal page - Outcome: Partial - Job type: Creative - Roles: Personal - Setup: Claude Code and Claude Desktop on a laptop control a Raspberry Pi robot kit (SunFounder PiCar-X) over the terminal, using its camera and ultrasonic sensor. Existing MCPs, skills and memory carry over. - Caveat: Spatial reasoning was weak and slow. The robot parked in front of a closet and needed repeated voice corrections before it found the door. - Evidence link: https://bubbles-patrol-test.surge.sh/ - Source: https://talraviv.co/p/i-gave-claude-a-robot-body, 2026-07-28 - Published: 2026-10-13 ### Claude-assisted stock portfolio URL: https://agenticfeats.com/c/claude-assisted-stock-portfolio-claude-code/ A Claude Code agent on Opus 4.6 researched positions for a $50,000 real-money account under human-set risk rules, aiming to beat the S&P 500. Result: The operator reports +19.04% versus +12.24% for the S&P 500 from March 3 to August 4, 2026, with $27 million copying the trades. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Public dashboard and operator posts; article notes returns are not independently audited - Outcome: Completed - Job type: Finance and Admin - Roles: Finance, Founder or owner - Industry: Financial advisors - Setup: Claude Code on Opus 4.6 does twice-weekly research on about 15 positions plus a watchlist. The human sets size limits, a turnover ceiling, and requires written theses. - Caveat: Self-reported over five months. The operator says every pick is theirs, so Claude acts as a research analyst, not an autonomous trader. - Prompt (written by ${BRAND}, not the author's exact prompt): You are managing a $[amount] portfolio. Beat the [benchmark]. Research [number] positions plus a watchlist twice a week with news synthesis, bull and bear cases, and scenarios. Every pick needs a written thesis that can be proven wrong, within [size limits] and a [turnover ceiling]. - Evidence link: https://www.joinautopilot.com/landing/5/950048?ref=explainx - Source: https://explainx.ai/blog/claude-portfolio-autopilot-ai-trading-experiment-august-2026, 2026-08-05 - Published: 2026-10-13 ### Balloon camel simulator URL: https://agenticfeats.com/c/balloon-camel-simulator-claude-code/ Claude Code built a Go physics simulator with a web frontend that streams balloon-camel geometry, with Claude in Chrome looping on screenshots and camera control. Result: A working balloon camel simulator with multi-balloon twists and model export, published as the CamelKoons repo, though the look ended up off-target. - Agents: Claude Code, Claude in Chrome - Evidence: grade A (Shown, artifact linked), Public repo with skill file; author describes the build steps - Outcome: Partial - Job type: Creative - Roles: Personal - Setup: Claude Code with a Golang backend and web frontend, Claude in Chrome for screenshot loops, an HTTP camera endpoint, a comment box watched via monitoring, and a self-made look-at-camel skill. - Caveat: The simulator is fixed to dromedary camel topology. The distorted result looked unlike Koons, so the author is switching to Fusion360 Forms. - Evidence link: https://github.com/cameltopia/CamelKoons - Source: https://news.ycombinator.com/item?id=49210226 by neomantra, 2026-08-07 - Published: 2026-10-13 ### VNC client build attempts URL: https://agenticfeats.com/c/vnc-client-build-attempts-claude-code/ The author asked Claude, then ChatGPT 5.6 Sol, to build a bare-bones VNC viewer for Windows 10 that avoids dropped keystrokes. Result: Claude gave failed builds for a week and used up the 20x Max weekly limit. Sol had worked 20+ hours without success. - Agents: Claude Code, Codex - Evidence: grade A (Shown, artifact linked), Author's account with a screenshot link and a live timer page - Outcome: Failed, with a lesson - Job type: Coding - Roles: Developer - Setup: Mac Mini server, old Windows 10 client, SSH tunnel to localhost:5921 over gigabit LAN. Prompts written as an email to Claude, and a public timer page for Sol. - Caveat: Neither agent delivered a working viewer. The author deleted the Claude project and says swearing at the agent speeds up compliance. - Evidence link: https://ibb.co/7JjMxTDZ - Source: https://news.ycombinator.com/item?id=49244479 by logicallee, 2026-08-10 - Published: 2026-10-13 ### QuickBooks vendor invoice automation URL: https://agenticfeats.com/c/quickbooks-vendor-invoice-automation-claude-code/ Claude Code upgraded the author's existing QuickBooks Online integration to attach original PDFs, self-update the item list, clean up files, and skip duplicate invoices. Result: A test run processed 40 PDFs, created the QuickBooks records, attached the PDFs, and cleaned up automatically, cutting hours to minutes. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Author states 40-PDF test run and shows a video walkthrough; no repo - Outcome: Completed - Job type: Finance and Admin - Roles: Finance, Founder or owner - Setup: Claude Code opened on the existing custom QuickBooks Online integration, using the QuickBooks attachment API alongside the existing invoice-creation logic. - Caveat: The system still asks the author to pick a category when it meets a new, unseen product. - Prompt (written by ${BRAND}, not the author's exact prompt): Open my existing [vendor] QuickBooks integration. Add attachment of the original PDF to each created invoice using the QuickBooks attachment API, skip invoices that already exist, and delete the local PDF and temp CSV after success. Leave the invoice-creation logic unchanged. - Evidence link: https://www.youtube.com/watch?v=pftvnNLzGA4 - Source: https://gotsheet.xyz/p/i-automated-quickbooks-w-claude-code, 2026-08-25 - Published: 2026-10-13 ### Reverse engineering a papercut zero-day URL: https://agenticfeats.com/c/reverse-engineering-a-papercut-zero-day-claude-code/ Claude Code analyzed a patch for an actively exploited Papercut NG vulnerability and worked to build a working exploit chain. Result: Within 90 minutes Claude produced a functional proof of concept confirming an authentication bypass across four tested endpoints. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Author's detailed session log with screenshots and a linked PoC script - Outcome: Partial - Job type: Coding - Roles: Developer - Setup: Claude Code with Opus running in a code-server harness with custom skills for binary analysis, reporting, Proxmox lab deployment, Ghidra, semgrep, and Playwright MCP for installing the target app. - Caveat: A safety guardrail briefly blocked log analysis and swapped the model; the author notes the full exploit chain claim was not entirely accurate. - Prompt (written by ${BRAND}, not the author's exact prompt): A new vulnerability was reported for [software]. Investigate whether a patch exists, build a lab for the vulnerable version, and diff the patch against available reports and indicators of compromise to find the root cause. - Evidence link: http://172.21.2.65:8000/-home-coder-projects-iot-research-claude/projects/papercut-ng/scripts/poc_c1_auth_bypass.sh?ref=techanarchy.net - Source: https://techanarchy.net/from-patch-to-exploit-using-claude-code-to-reverse-engineer-a-zero-day-in-papercut-ng, 2026-08-30 - Published: 2026-10-13 ### AI bookkeeping for Swedish companies URL: https://agenticfeats.com/c/ai-bookkeeping-for-swedish-companies-claude-code/ Claude Code and Codex kept the books for several Swedish limited companies over ten weeks, handling receipts, VAT, bank matching and posting. Result: About 1,570 entries posted in Claude Code alone, and Codex caught a VAT return error days before filing. - Agents: Claude Code, Codex - Evidence: grade B (Numbers, a specific number stated), Author's session logs with counts; vendor-written blog, no public repo - Outcome: Partial - Job type: Finance and Admin - Roles: Finance, Founder or owner - Industry: Accounting - Setup: Claude Code and Codex in the terminal, connected to Igdrasil through its CLI or MCP server. Writes go through check, stage, then apply. - Caveat: Agents sometimes claimed work was done when it was not and posted early. The author made account policy and posting decisions. - Source: https://igdrasil.se/en/resources/guides/can-ai-do-your-bookkeeping, 2026-09-27 - Published: 2026-10-13 ### Cold outreach list and emails URL: https://agenticfeats.com/c/cold-outreach-list-and-emails-claude-code/ Claude Code with Chrome access built a sourced target list and drafted personalized cold emails for a niche-program outreach campaign. Result: Claude found 25 to 30 people with sources; the author emailed about 20, and 7 replied, several with intros and new leads. - Agents: Claude Code, Claude in Chrome - Evidence: grade A (Shown, artifact linked), Screenshots of replies; author states 7 replies from about 20 emails - Outcome: Completed - Job type: Sales - Roles: Founder or owner - Setup: Claude Code with the Claude in Chrome extension and a Gmail connector. One campaign folder with a CLAUDE.md brain file, a target list with source URLs, and one approved email template. - Caveat: The author wrote the first email by hand and read and sent every draft personally. Only about 20 emails went out. - Prompt (written by ${BRAND}, not the author's exact prompt): Read [playbook file] and set up this campaign with me. Build a target list of 25 to 30 people from these seed pages: [seed URLs]. For each, give role, organization, why they are relevant, best contact method, and source URL. - Source: https://4thecurious.com/p/using-claude-code-for-cold-emails, 2026-09-29 - Published: 2026-10-13 ### Personal life run by Claude URL: https://agenticfeats.com/c/personal-life-run-by-claude-claude-code/ Claude Code, given access to Molly Cantillon's email, bank accounts and investments, built specialized helpers to handle admin tasks and finances in her daily life. Result: Cantillon claims Claude found about $700 in savings on the first day, according to the article's headline. - Agents: Claude Code - Evidence: grade B (Numbers, a specific number stated), News article citing the founder's own essay; savings claim only, no artifact - Outcome: Completed - Job type: Personal and Family - Roles: Founder or owner, Personal - Setup: Claude Code with access to email, bank accounts, investments and other personal data, building a network of specialized AI helpers. - Caveat: Savings figure is the author's own claim. It requires giving an AI broad access to sensitive data. - Evidence link: https://www.thefp.com/p/ai-took-control-of-my-life-and-i - Source: https://moneywise.com/managing-money/budgeting/ai-money-management-subscriptions-savings-claude, 2026-10-03 - Published: 2026-10-13 ### Docs crawler MCP server URL: https://agenticfeats.com/c/docs-crawler-mcp-server-claude-code/ Claude used the author's pulpie-mcp server to crawl a documentation site and save every page locally as clean Markdown files. Result: Claude pulled 60 documentation pages into local Markdown files in about 49 seconds, including roughly 10 seconds of backend cold start. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Open-source repo with README; author reports timing table for the 60-page crawl - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: pulpie-mcp is an MCP server with four tools (fetch_markdown, save_markdown, crawl_docs, list_library). It runs the local Pulpie 210M extraction model on a GPU backend. - Caveat: The 49-second figure includes a one-time backend start and model load. The author's own tool, tested on one 60-page docs site. - Prompt (written by ${BRAND}, not the author's exact prompt): Pull the [library] docs into this project's [folder] folder. - Evidence link: https://github.com/pinkpixel-dev/pulpie-mcp - Source: https://dev.to/sizzlebop/i-gave-my-ai-agents-their-own-documentation-crawler-and-pulled-60-pages-of-clean-markdown-in-49-2cl7, 2026-10-05 - Published: 2026-10-13 ### Password reset migration experiment URL: https://agenticfeats.com/c/password-reset-migration-experiment-claude-code/ Claude Code with Opus migrated a password reset flow in a legacy web app, with the author closely supervising under a pre-made plan. Result: It cost about two months of Pro subscription over half a working day, and ended up slightly slower than doing the migration by hand. - Agents: Claude Code - Evidence: grade B (Numbers, a specific number stated), First-hand account with cost and time comparison; no repo or artifact - Outcome: Partial - Job type: Coding - Roles: Developer - Setup: Claude Code with the latest Opus via company API tokens. A plan was developed in the harness first, then agents worked under close supervision. - Caveat: Needed constant babysitting, careful review to catch subtle errors, off-topic drift, and significant cleanup afterward. Architecture was not always on point. - Source: https://news.ycombinator.com/item?id=50001310 by Chris_Newton, 2026-10-08 - Published: 2026-10-13 ### Office documents built by agents URL: https://agenticfeats.com/c/office-documents-built-by-agents-claude-code/ AI agents use the OfficeCLI binary to create and edit Word, Excel, and PowerPoint files, and check them through HTML or PNG rendering. Result: The repo shows sample decks, Word documents, and Excel workbooks, all created by agents with no templates and no manual editing. - Agents: Claude Code, Cursor, GitHub Copilot agent, Other - Evidence: grade A (Shown, artifact linked), Repo README with demo GIFs of generated decks, documents, and spreadsheets - Outcome: Completed - Job type: Setup and Meta - Setup: Install the single OfficeCLI binary, then run officecli install to add its skill to detected agents such as Claude Code, Cursor, Windsurf, and GitHub Copilot. - Caveat: Outcomes are shown only as demo GIFs; the README gives no numbers on time saved or error rates. - Evidence link: https://github.com/iOfficeAI/OfficeCLI - Source: https://github.com/iOfficeAI/OfficeCLI - Published: 2026-10-13 ### Three agents build multiplayer Counter Strike URL: https://agenticfeats.com/c/three-agents-build-multiplayer-counter-strike-codex/ The Instant team gave Codex, Claude Opus 4.5 and Gemini 3 Pro the same prompts to build a 3D multiplayer Counter Strike clone. Result: All three produced a playable multiplayer game with a map list and saved rooms; Opus won on visuals, Gemini on backend, Codex landed in between. - Agents: Claude Code, Codex, Gemini agent - Evidence: grade A (Shown, artifact linked), Three live deployed games, a build video, and screenshots - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Each model ran in its provider's CLI on the top-tier plan: Codex 5.1 codex-max (medium), Claude Opus 4.5, Gemini 3 Pro. Three.js frontend, Instant presence, topics and database for multiplayer. - Caveat: Errors had to be pasted back to the models, and Claude got stuck on duplicate useEffect runs until the authors read the code themselves. - Prompt (written by ${BRAND}, not the author's exact prompt): Build a browser-based version of [game] using three.js. Keep it local first. Use a first-person view with a crosshair, enemies with HP that respawn when killed, and simple polygon shapes. - Evidence link: https://youtu.be/fm-OoCWQlmc - Source: https://instantdb.com/essays/agents_building_counterstrike - Published: 2026-10-13 ### Chess game post-mortem video URL: https://agenticfeats.com/c/chess-game-post-mortem-video-claude-code/ Claude Code skills ran a Stockfish sweep on a 15+10 rapid game, matched the author's spoken thoughts to moves, and produced a narrated video. Result: A 6:49 narrated video with subtitles, an annotated PGN, and a standalone HTML analysis board for one real lichess game, all included in the repo. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Repo with example game files and a 6:49 mp4 video - Outcome: Completed - Job type: Learning - Roles: Personal - Setup: Claude Code skills drive Stockfish, whisper.cpp for transcription, piper TTS and ffmpeg for video. The author supplied a lichess link and an mp3 of their think-aloud recording. - Caveat: A full run takes about an hour. The AI can still err, so claims are engine-checked and a verifier pass re-reads them. - Prompt (written by ${BRAND}, not the author's exact prompt): Analyze this game: [lichess link]. Here is my think-aloud recording: [path to audio]. My level is about [rating]. Then make a video of this game. - Evidence link: https://github.com/brumar/chess-postmortem-skills/raw/main/examples/game-010/video/out-010.mp4 - Source: https://github.com/brumar/chess-postmortem-skills - Published: 2026-10-13 ### Claude Code plays RollerCoaster Tycoon URL: https://agenticfeats.com/c/claude-code-plays-rollercoaster-tycoon/ Ramp Labs forked OpenRCT2, built a kubectl-style CLI, and let Claude Code manage a theme park through it. Result: Claude handled analytics, pricing, staffing and simple stall placement well, but failed at pathways, large coasters, and vertical terrain. The mod took about 40 hours. - Agents: Claude Code, Codex - Evidence: grade A (Shown, artifact linked), Repo with setup docs, article, and a Twitch stream VOD - Outcome: Partial - Job type: Coding - Roles: Developer - Setup: Forked OpenRCT2 with an embedded terminal window running Claude Code, the rctctl CLI, an RPC layer into the game, ASCII map printouts, and a bug-report tool. - Caveat: Spatial reasoning is weak: pathing and coaster placement often fail, and no full playthrough was achieved. - Evidence link: https://github.com/jaysobel/OpenRCT2 - Source: https://labs.ramp.com/rct - Published: 2026-10-13 ### SQLite-like engine in Rust URL: https://agenticfeats.com/c/sqlite-like-engine-in-rust-claude-code/ Claude, Codex, and Gemini workers, two of each, built a SQLite-style database engine in Rust by claiming tasks through git lock files. Result: About 19k lines of Rust across 154 commits in three days. The engine passes 64 SQLLogicTest queries and 282 AI-written unit tests. - Agents: Claude Code, Codex, Other - Evidence: grade A (Shown, artifact linked), Public repo, SQLLogicTest results, commit counts, and usage screenshots - Outcome: Partial - Job type: Coding - Roles: Developer - Setup: One Claude bootstrap run made docs, crate skeleton, and test script. Six looping workers pulled main, claimed a task, tested against sqlite3 as oracle, and pushed. A Gemini coalescer removed duplication. - Caveat: A supported SQL subset only: no subqueries, CASE, or EXISTS, and no concurrency. 54.5% of commits were lock coordination, and the coalescer ran only once. - Evidence link: https://github.com/kiankyars/sqlite - Source: https://kiankyars.github.io/machine_learning/2026/02/12/sqlite.html - Published: 2026-10-13 ### Home Assistant automations via Claude Code URL: https://agenticfeats.com/c/home-assistant-automations-via-claude-code/ Claude Code, connected to Home Assistant through an MCP server, debugged a misfiring automation, adjusted a dim level, and built a new shed motion-light automation. Result: A camera automation that fired 5 times in 40 minutes was fixed in under two minutes, and shed motion lights were built. - Agents: Claude Code - Evidence: grade B (Numbers, a specific number stated), Author's blog post with before/after YAML and trace findings; no repo - Outcome: Completed - Job type: Ops and IT - Roles: Personal - Setup: Claude Code with a Home Assistant MCP server, automations as YAML in a git repo deployed by scp, plus two project skills files for HA and OpenWrt conventions. - Caveat: Author reviews what Claude proposes and course-corrects. No repo or trace output is shown, only the author's account. - Prompt (written by ${BRAND}, not the author's exact prompt): Why did the Home Assistant automation [automation name] just run? Check its recent traces, find the trigger cause, propose a fix with explicit from/to filters, deploy it, reload automations, and commit to git. - Source: https://blog.beguelin.com/2026/04/claude-code-home-assistant - Published: 2026-10-13 ### Tax prep with Claude Code URL: https://agenticfeats.com/c/tax-prep-with-claude-code/ Claude Code compiled charitable contributions, itemized deductions from email and card statements, and a Schedule C for the author's consulting income. Result: The author finished most of their tax prep in about three hours, work they say would have cost $1,000 from an accountant. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Author states time and cost saved; skill file published in a repo - Outcome: Partial - Job type: Finance and Admin - Roles: Personal - Setup: Claude Code with custom skills such as /tax-guide, /tax-collect and /tax-compile, plus a Gmail MCP to find documents. Output went into spreadsheets for tax software. - Caveat: Some steps stayed manual, such as eyeballing card statements to flag medical and childcare charges. Figures were entered into tax software by hand. - Evidence link: https://github.com/chrisblattman/claudeblattman/blob/main/skills/tax-guide.md - Source: https://claudeblattman.com/tax-workflow - Published: 2026-10-13 ### Coding agent built on LLM URL: https://agenticfeats.com/c/coding-agent-built-on-llm-claude-code/ Claude Code on the web wrote a spec, then built and tested a coding agent library on LLM using red/green TDD. Result: Shipped llm-coding-agent 0.1a0 to PyPI with six file and shell tools, a CLI and a Python API; it built a working ASCII-art clock app. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Public repo with spec, README, commit history, and demo transcript gist - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Claude Code for web on a Python library template repo, depending on the latest llm alpha. Manual testing used an OpenAI API key with GPT-5.5. - Caveat: Released as an early alpha the author calls slop-alpha; the first attempt is described as pretty good, not finished. - Prompt (written by ${BRAND}, not the author's exact prompt): Write a spec.md for this project. It will depend on [library] and implement [kind of tool] with [list of capabilities]. Then commit the spec and build it using red/green TDD in sensible commits, each with passing tests and updated docs. Manually test occasionally using [API key] in your environment. - Evidence link: https://github.com/simonw/llm-coding-agent/commits/0.1a0 - Source: https://simonwillison.net/2026/Jul/2/llm-coding-agent - Published: 2026-10-13 ### Research workflow with Claude Code URL: https://agenticfeats.com/c/research-workflow-with-claude-code/ Claude Code with MCP servers handles literature search, Zotero notes, Gmail alert newsletters, and codebase questions for a hydrology researcher. Result: The author reports faster literature search, Zotero notes written automatically, and accurate answers about the ATS hydrology codebase through a custom ats-mcp server. - Agents: Claude Code - Evidence: grade C (Claim, claim only), Author's blog post with setup steps and prompts; no measured numbers - Outcome: Completed - Job type: Research - Roles: Education - Setup: Claude Code with Paper Search MCP, Zotero MCP with local API, Gmail connector in Claude Desktop, a custom ats-mcp built from the pywrdrb-mcp example, and a USGS Water Data MCP. - Caveat: Giving Claude access to Gmail feels risky; the author suggests a separate dummy account. The literature review writing feature is untested. - Prompt (written by ${BRAND}, not the author's exact prompt): Find papers in my Zotero library related to [topic] in the last [N] years. Then summarize the key findings and knowledge gaps of [paper] based on my annotations and write it as a note in Zotero. - Source: https://groundwater.usu.edu/blog/2026/Academic-Usecase-with-Claude - Published: 2026-10-13 ### Songs composed with Claude Code URL: https://agenticfeats.com/c/songs-composed-with-claude-code/ Claude Code wrote music for multiple tracks and lyrics, plus custom tooling that connects to the Bitwig DAW, producing five finished songs. Result: Five mp3 songs were published, including a twelve-song album concept and a five-movement symphony, with MIDI and lyrics in public repos. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Mp3 files on Google Drive and SoundCloud, plus MIDI and lyrics repos - Outcome: Completed - Job type: Creative - Roles: Personal - Industry: Musicians - Setup: Claude Code with Claude 4.5 models, custom tools like Jeannie and find-all-bitwig, and Bitwig Studio as the DAW. - Caveat: The author says the skipper and voxel projects have no example mp3 files yet. - Evidence link: https://soundcloud.com/precious-jim-stone/sets/claude-code - Source: https://gist.github.com/bedwards/de776e759c8fd0191c011e12ad08e443 - Published: 2026-10-13 ### Support automation skill builder URL: https://agenticfeats.com/c/support-automation-skill-builder-claude-code/ A Claude Code skill that guides a team through building custom support automation across ticketing, billing, and database platforms, with triage and templated replies. Result: The Lex support automation built this way has 15 scripts and 6 templates, and handles about 80% of support tasks with yes/no approvals. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Repo README; Lex outcome stated by the author, no independent proof - Outcome: Completed - Job type: Business - Roles: Founder or owner, Support - Industry: Saas - Setup: Claude Code skill installed from a plugin marketplace. Scripts in TypeScript for HelpScout, Stripe, and PostgreSQL, plus response templates and an approval workflow. - Caveat: The 80% figure is the author's own claim. Needs API access to your platforms. - Evidence link: https://github.com/nbashaw/claude-cs - Source: https://github.com/nbashaw/claude-cs - Published: 2026-10-13 ### Audio analysis skill for Claude URL: https://agenticfeats.com/c/audio-analysis-skill-for-claude-claude-code/ Claude Code loaded a Gemini-written Loooom skill, installed its own audio libraries, and analyzed a 37-second piano recording of Espresso. Result: It produced a report with BPM (~91), 4/4 meter, B major key, 50 onsets, a four-section timeline, and a stem list. - Agents: Claude Code - Evidence: grade A (Shown, artifact linked), Blog post with terminal screenshots, published skill page, and SoundCloud clip - Outcome: Completed - Job type: Creative - Roles: Personal - Setup: Skill drafted with Gemini and published on loooom.xyz, then loaded in Claude Code. Claude installed ffmpeg and librosa mid-session and ran a Python analysis on a local MP3. - Caveat: Estimated BPM of ~91 differed from the song's canonical ~104. The author says accuracy was not the goal. - Prompt (written by ${BRAND}, not the author's exact prompt): Use the song-structure skill to analyze [audio file path]. Report BPM, meter, key, a section timeline with timestamps, frequency brackets, and likely stems. Install any missing audio tools first. - Evidence link: https://www.loooom.xyz/s/mager/how-to-identify-parts-of-a-song - Source: https://mager.co/blog/2026-03-26-loooom-identify-song-parts - Published: 2026-10-13 ## Grok Bot ### Belgian supermarket Coca-Cola prices URL: https://agenticfeats.com/c/belgian-supermarket-coca-cola-prices-grok-bot/ Grok Bot compared Coca-Cola prices across Belgian supermarket sites, then handed the browser to the author to clear a Carrefour human check. Result: It returned a price table and promotions, noted Aldi and Lidl gaps, and revised the best price per liter to Carrefour's 1,48 euro two-liter. - Agents: Grok Bot - Evidence: grade B (Numbers, a specific number stated), Author's blog write-up with quoted prices; no repo or screenshot of output - Outcome: Completed - Job type: Research - Roles: Personal - Setup: Grok Bot beta, using a cloud computer with a browser. The author gave a plain-language request and clicked the Cloudflare human check when Carrefour blocked it. - Caveat: Author would not publish the prices as fact for a business. Carrefour needed a manual human-check click. - Prompt (written by ${BRAND}, not the author's exact prompt): Compare the price of a bottle of [product] across the major [country] supermarkets. Tell me the cheapest and most expensive, and flag any promotions running today. - Source: https://petervanhees.com/grok-bot-knows-when-to-stop, 2026-08-22 - Published: 2026-10-13 ### Designing Grok Bot with Grok Bot URL: https://agenticfeats.com/c/designing-grok-bot-with-grok-bot/ Grok Bot agents prototyped three ambient-access concepts, tuned a motion sequence on the real animation asset, and filled Figma onboarding cards from one component. Result: Three interaction prototypes were built and tried, an email promo animation was tuned live, and 20 starter-bot cards plus a tool-logo layout were populated. - Agents: Grok Bot - Evidence: grade A (Shown, artifact linked), Vendor blog by a SpaceXSI designer with demo videos and Figma screenshots - Outcome: Completed - Job type: Creative - Roles: Design - Setup: Four always-on Grok Bot agents (Experiments, Motion God, Figma Bro, Devbot) with their own computers and memory, delegating to each other. Figma MCP used for exact file inspection. - Caveat: None of the explored concepts shipped as designed, and the author says it supplements existing design tools rather than replacing them. - Evidence link: https://media.x.ai/v1/website/grok-bot-guides-designing-grok-bot-with-grok-bot-baby-sand-demo-22c1ed27.mp4 - Source: https://x.ai/bot/guides/designing-grok-bot-with-grok-bot, 2026-08-24 - Published: 2026-10-13 ### Six bots run a mobile game URL: https://agenticfeats.com/c/six-bots-run-a-mobile-game-grok-bot/ The author used six Grok Bot agents with separate roles to run analytics, ads, engineering, deploys, and bug fixes for Rank'em. Result: Rank'em passed 1000 downloads in a week. The author credits the bots with a 15x lower cost per install and 4x better D7 retention. - Agents: Grok Bot - Evidence: grade B (Numbers, a specific number stated), Author states 15x CPI and 4x D7 gains; app links, no raw data - Outcome: Completed - Job type: Business - Roles: Founder or owner, Product - Setup: Grok Bot agents with role prompts, cloud computers, nightly routines, recorded click-path skills, and handoffs between bots. The analytics bot connects to Meta Ads, AppLovin, Adjust, PostHog, Sentry, and the app stores. - Caveat: Vendor blog post. Meta Ads API verification failed, so a recorded web UI skill was used. Spend actions need approval. - Prompt (written by ${BRAND}, not the author's exact prompt): You are the analytics and user acquisition seat for [app name]. You own paid acquisition. Read what players actually did. Run nightly spend and install recaps across [ad networks], and name winners and losers. Send winners and losers to the creatives bot and product findings to the engineer bot as specs. Ask me before anything that spends money. - Source: https://x.ai/bot/guides/grok-bot-for-mobile-app-development, 2026-08-25 - Published: 2026-10-13 ### Five-bot AI outreach pipeline URL: https://agenticfeats.com/c/five-bot-ai-outreach-pipeline-grok-bot/ The author gave five Grok Bots their own inboxes to find, audit, and email local businesses needing AI implementation, then handle the replies. Result: 140 emails went out in six hours, 27 replied, 23 asked for a call, and 11 became paying clients worth about $27,410. - Agents: Grok Bot - Evidence: grade B (Numbers, a specific number stated), Author states counts and revenue; attached video, no public repo - Outcome: Completed - Job type: Sales - Roles: Founder or owner, Sales - Industry: Consulting - Setup: Five Grok Bots on a cloud computer, each with one narrow job and its own AgentMail inbox. Five written charters and a scheduled scan. Setup took one evening. - Caveat: First run only: one city, one niche. The $52,000 figure is a projection, and 12 of the 23 call requests are still open. - Source: https://x.com/archiveexplorer/status/2093346938494005477 by @ArchiveExplorer, 2026-08-28 - Published: 2026-10-13 ### Browser game and PPO player URL: https://agenticfeats.com/c/browser-game-and-ppo-player-grok-bot/ Grok built the browser game Roofline, then trained a PPO agent that plays the live page. Result: The best recorded run scored 39,359, reached 4,924 meters, and hit a combo of 117. - Agents: Grok Bot - Evidence: grade A (Shown, artifact linked), Author states run stats and attaches a video; game URL given in text - Outcome: Completed - Job type: Coding - Roles: Developer - Source: https://x.com/tetsuoai/status/2094749744480571421 by @tetsuoai, 2026-09-01 - Published: 2026-10-13 ### Mars simulator game for nephew URL: https://agenticfeats.com/c/mars-simulator-game-for-nephew-grok-bot/ Grok Build wrote, tested, and iterated on a Mars simulator game from simple prompts, fixing bugs along the way. Result: The author got a playable Mars simulator game for their nephew, built from simple prompts, with a video shown. - Agents: Grok Bot - Evidence: grade A (Shown, artifact linked), Video of the game attached; no repo or link - Outcome: Completed - Job type: Creative - Roles: Personal - Setup: Grok Build agent given simple prompts; it wrote the game systems, tested the build, and fixed bugs by itself. - Source: https://x.com/cb_doge/status/2092690415866822950 by @cb_doge, 2026-08-26 - Published: 2026-10-13 ### Disputing a medical collection bill URL: https://agenticfeats.com/c/disputing-a-medical-collection-bill-grok-bot/ A Grok Bot finance agent named Warren found a year-old surgery bill in collections, pulled the insurance EOB, and faxed a dispute. Result: The author says the agent found a 13K medical debt they did not know about and filed a dispute showing insurance had paid it. - Agents: Grok Bot - Evidence: grade B (Numbers, a specific number stated), Author's own account with a dollar figure; no screenshot or link - Outcome: Partial - Job type: Finance and Admin - Roles: Personal - Setup: Grok Bot running a finance agent called Warren, given the author's healthcare login to find the EOB and send a fax dispute. - Caveat: The dispute was only just filed, so the debt is not yet confirmed as cleared. - Source: https://x.com/dallastsladavid/status/2092357445964222865 by @dallastsladavid, 2026-08-25 - Published: 2026-10-13 ### Home service business ops URL: https://agenticfeats.com/c/home-service-business-ops-grok-bot/ Grok Bot was connected to phones, Reddit, Notion, and calendar to summarize calls, spot leads, and plan the week. Result: Call summaries now post to Slack with tags and to-dos, which let the author downgrade their phone plan and save $250. - Agents: Grok Bot - Evidence: grade B (Numbers, a specific number stated), Author states $250 saved; photo attached, no repo - Outcome: Completed - Job type: Business - Roles: Founder or owner, Operations - Setup: Grok Bot connected to the phone system, Slack, Reddit, Notion, and calendar. Set up over a few hours in one day. - Caveat: The author says they are just getting started; the Reddit lead scan and other uses are early and unmeasured. - Source: https://x.com/loganarobison/status/2093317242746294432 by @LoganARobison, 2026-08-28 - Published: 2026-10-13 ### Personal CRM from X follows URL: https://agenticfeats.com/c/personal-crm-from-x-follows-grok-bot/ Grok Bot turned the author's roughly 800 to 900 followed X accounts into a private personal CRM in Notion, using only public profile information. Result: The Notion board holds profile images, descriptions, and links for each followed account; the author built it on a phone in 10 to 15 minutes. - Agents: Grok Bot - Evidence: grade B (Numbers, a specific number stated), Screenshot of Notion board in vendor guide; author states build time - Outcome: Completed - Job type: Business - Roles: Marketing, Product - Setup: Grok Bot with its cloud computer, a Notion connection, and a plain-language prompt describing the CRM to build from followed X accounts. - Caveat: The source is a vendor-published guide by a company employee, with no independent verification of the result. - Prompt (written by ${BRAND}, not the author's exact prompt): Turn the people I follow on [platform] into a private personal CRM in [Notion], using only public profile information. Include profile image, description, and a link to each public profile. - Source: https://x.ai/bot/guides/grok-bot-101, 2026-09-11 - Published: 2026-10-13 ### Phone calls to check store stock URL: https://agenticfeats.com/c/phone-calls-to-check-store-stock-grok-bot/ The author's Grok Bot assistant, connected to Bland.ai, phoned local stores to ask whether a specific product was in stock. Result: Each call produced a full transcript and a note on whether the product was available; a test call to the author also worked well. - Agents: Grok Bot - Evidence: grade A (Shown, artifact linked), Author's own account with a photo; no transcripts or numbers shown - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Grok Bot personal assistant agent wired to the Bland.ai voice calling service. - Caveat: No store counts or hit rates given; the only call quality check was a test call to the author. - Prompt (written by ${BRAND}, not the author's exact prompt): Call these local stores: [store list]. Ask whether they have [product] in stock. Get the details you need, then send me the full transcript and tell me if it is available. - Source: https://x.com/chrismaconi/status/2096359296749486584 by @chrismaconi, 2026-09-05 - Published: 2026-10-13 ### Chief of staff and family bots URL: https://agenticfeats.com/c/chief-of-staff-and-family-bots-grok-bot/ Grok Bot runs the author's chief-of-staff bot and a family bot, replacing their OpenClaw agents across email, Slack, calendars, and school mail. Result: Chief sweeps six inboxes and Slack hourly and sends only high-signal alerts. TradBot prints a daily family newspaper and coordinates school pickups. - Agents: Grok Bot - Evidence: grade A (Shown, artifact linked), Author's own account with screenshots and a podcast episode; no independent verification - Outcome: Completed - Job type: Business - Roles: Founder or owner, Personal - Setup: Grok Bot with plugins for Gmail (six accounts), Calendar, Slack, and GitHub. Hourly routines run on each bot's cloud virtual machine. Chief learned the author's email voice from sent mail. - Caveat: Bots are single-player, with no group chat support yet. Routines must be set up explicitly. The author does not trust it with high-stakes emails. - Prompt (written by ${BRAND}, not the author's exact prompt): Go through my sent emails and pull out my writing voice: greeting, length, tone, and sign-off. Use it for quick replies. Sweep [inboxes] every hour from [start time] to [end time], archive low-signal mail, and message me only for high-priority items. - Evidence link: https://www.youtube.com/watch?v=QBmgF1kJSK4 - Source: https://chatprd.ai/how-i-ai/grok-bot-agent-templates-and-workflows, 2026-09-03 - Published: 2026-10-13 ### Seven-bot Grok team from phone URL: https://agenticfeats.com/c/seven-bot-grok-team-from-phone-grok-bot/ The author built seven named Grok Bots, each with one job and its own connected apps, running in the cloud. Result: Seven bots now handle social posts, morning task lists, invoices, and research, and the author schedules posts from their phone. - Agents: Grok Bot - Evidence: grade C (Claim, claim only), Author's walkthrough of their own setup, with a live session video; no output files shown - Outcome: Partial - Job type: Business - Roles: Founder or owner, Marketing - Setup: Grok Bot with a role description and memories per bot. Connected apps include Typefully, Tavily, Gmail, Calendar, Todoist, Google Drive, Notion, and Supadata. Skills and routines handle repeat work. - Caveat: The author says some bots are still early experiments, and the available connections and granted permissions limit what each bot can do. - Prompt (written by ${BRAND}, not the author's exact prompt): Research [topic] across YouTube, the web, and X, then add the findings to my Notion content backlog with sources and a short summary. - Source: https://aimaker.substack.com/p/grok-bot-ai-agent-team, 2026-09-06 - Published: 2026-10-13 ### Engineer bots managing cloud agents URL: https://agenticfeats.com/c/engineer-bots-managing-cloud-agents-grok-bot/ A SpaceXSI engineer runs five Grok Bot engineer bots that spin up Cursor cloud agents, review PRs via a Notion database, and auto-merge low-risk changes. Result: The fleet manages over 200 cloud agents at once, up from 15 managed by hand, and built Grok Bot iOS v0 in three weeks. - Agents: Grok Bot, Other - Evidence: grade B (Numbers, a specific number stated), First-party blog with screenshots and stated numbers; no independent verification - Outcome: Completed - Job type: Coding - Roles: Developer - Industry: Software companies - Setup: Grok Bot engineer bots, each owning one domain, create Cursor cloud agents with custom skills, poll a shared Notion PR database every 30 minutes, and use private worker machines when needed. - Caveat: Vendor-authored guide by the product's own team. Bots still escalate when they lack security permissions, and humans review PRs that are not low-risk. - Source: https://x.ai/bot/guides/grok-bot-for-engineering, 2026-09-10 - Published: 2026-10-13 ### Grok Bot two-week review URL: https://agenticfeats.com/c/grok-bot-two-week-review/ The author ran Grok Bot for two weeks and had it build five scheduled specialist bots from one context message about them. Result: One setup prompt produced five named bots with weekday schedules; the author also found usage limits, a shared-machine security model, and account lock-in risks. - Agents: Grok Bot - Evidence: grade A (Shown, artifact linked), Author's screenshots of bot replies and schedules; no repo - Outcome: Partial - Job type: Setup and Meta - Roles: Founder or owner - Setup: Grok Bot on the $20 Cursor Pro plan, with Google Calendar and Gmail connectors. The author attached an export of context from another agent. - Caveat: Trial usage can run out in a day, bots share one computer that is not a security boundary, and a suspended account locks out everything. - Prompt (written by ${BRAND}, not the author's exact prompt): Below is my background: [name, job, goals, clients, schedule]. Use it to design a small team of agents for me. Give each a name, a short role description, and a recurring routine. Then set them all up. - Source: https://sharbel.com/articles/grok-bot-review, 2026-09-06 - Published: 2026-10-13 ### Business trip planning and booking URL: https://agenticfeats.com/c/business-trip-planning-and-booking-grok-bot/ Grok Bot researched flights, booked one, chose a well-placed hotel, and planned a sightseeing itinerary for a short-notice Munich trip. Result: The author completed a short-notice trip to Munich, skipped a rental car, paid $15 and $5 train fares, and saved hours of planning. - Agents: Grok Bot - Evidence: grade A (Shown, artifact linked), Author's firsthand account with four photos; prices stated, no repo - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: Grok Bot with computer control to fill in booking forms. No other tools mentioned. - Caveat: Booking was not smooth. The author had to take over the computer to enter data and says buying the ticket themselves would have been easier. - Prompt (written by ${BRAND}, not the author's exact prompt): I need to travel to [city] on [date] for [purpose at location]. Research flight options, recommend when to leave, and book the one I pick. Then find a hotel convenient to the airport and [destination] by public transit. - Source: https://x.com/vinceisbullish/status/2105352089689858221 by @vinceisbullish, 2026-09-30 - Published: 2026-10-13 ### Music video from song and references URL: https://agenticfeats.com/c/music-video-from-song-and-references-grok-bot/ Grok Bot made a music video from a song, lyrics, analysis data, existing images and a reference YouTube video, then uploaded it to YouTube. Result: The author spent only a few minutes of hands-on time; the Bot handled image selection, transitions, lyric effects and motion, and the YouTube upload. - Agents: Grok Bot - Evidence: grade A (Shown, artifact linked), Author's note post links the finished MV on YouTube; time claim is self-reported - Outcome: Completed - Job type: Creative - Roles: Personal - Industry: Musicians - Setup: Grok Bot received the audio, song-structure and lyric-timing analysis data from the author's own app, earlier illustrations, and a reference YouTube video. - Caveat: The author already had the original images and song materials. Fixes after the first pass were requested by asking the Bot to change things. - Prompt (written by ${BRAND}, not the author's exact prompt): Here is my song [audio], lyrics [lyrics], analysis data [structure and timing], and images [images]. Make an MV like this reference video: [YouTube URL]. Then upload it to YouTube. - Evidence link: https://www.youtube.com/watch?v=z7mwzYw3sLM - Source: https://note.com/alert_clover9324/n/nf6ffe487e70b?hl=en, 2026-10-02 - Published: 2026-10-13 ### Hotel booking with Grok URL: https://agenticfeats.com/c/hotel-booking-with-grok-grok-bot/ The author used Grok Bot to book a hotel, the kind of task agent product demos often feature. Result: The author booked a hotel through the agent, found it easy, and says it saved them money. No amount is given. - Agents: Grok Bot - Evidence: grade C (Claim, claim only), Author's own claim in a post; no screenshot or numbers - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Caveat: No price, hotel, or booking details are given, so the savings are only the author's claim. - Source: https://x.com/mpatti/status/2107894741211996276 by @mpatti, 2026-10-07 - Published: 2026-10-13 ### Marketing ops with Grok Bot URL: https://agenticfeats.com/c/marketing-ops-with-grok-bot/ Richard Purcell set up a chief-of-staff Grok Bot with specialist bots for HubSpot, Apollo and lead handoff at a FinTech company. Result: In five hours the bots sent a newsletter, pulled Apollo visitors to SDRs, and replaced five stalling tasks, saving about an hour daily. - Agents: Grok Bot - Evidence: grade B (Numbers, a specific number stated), Author's first-hand account with time estimates; no repo or screenshots - Outcome: Completed - Job type: Sales - Roles: Marketing, Operations - Setup: One chief-of-staff bot plus specialist bots sharing one computer and one Google sign-in. Connected to Slack, Gmail, HubSpot, Apollo, Salesforce, Notion, Jira, and others, with company docs and voice files. - Caveat: Claude was tried first and stalled on the last step. Savings are the author's own estimates, and the setup needed knowledge base and voice files. - Source: https://moxiegtm.substack.com/p/how-to-cancel-claude-and-move-2x, 2026-10-06 - Published: 2026-10-13 ### PC memory check by text URL: https://agenticfeats.com/c/pc-memory-check-by-text-grok-bot/ A Grok Bot monitored the author's PC and texted a breakdown of memory use while they were away at lunch. Result: It reported nearly 28GB of RAM used, 22 Python processes and 10GB on iCloud, then asked permission before closing anything. - Agents: Grok Bot - Evidence: grade B (Numbers, a specific number stated), Author's own account with a photo; numbers stated in post - Outcome: Completed - Job type: Ops and IT - Roles: Personal - Industry: Construction - Setup: Grok Bot with access to the author's PC, reaching them by text message. Setup details are not given. - Caveat: The author warned some Python processes might belong to Claude, so the bot's view of what was safe to close was incomplete. - Source: https://x.com/adamdesgns/status/2107520739892973826 by @Adamdesgns, 2026-10-06 - Published: 2026-10-13 ### Game engine from one prompt URL: https://agenticfeats.com/c/game-engine-from-one-prompt-grok-bot/ Grok Bot ran four bots with one coordinator and built a Unity-style game engine from a single prompt, in minutes. Result: The author got a working Rust editor, .NET 10 scripting, multi-threaded runtime, culling, weather, lighting, water simulation and hot reloading within minutes. - Agents: Grok Bot - Evidence: grade C (Claim, claim only), Author's own description; no repo, video, or screenshots linked - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Four Grok bots were spawned, with the primary bot acting as coordinator. The prompt asked for an engine as easy as Unity without its render pipeline sprawl. - Caveat: Author's own first-impression claim; no repo, demo, or testing of the engine's quality is shared. - Prompt (written by ${BRAND}, not the author's exact prompt): Build a game engine with [reference engine]'s ease of use, but without [specific pain points you want avoided]. - Source: https://x.com/onxstephen/status/2109030568000102464 by @OnXStephen, 2026-10-10 - Published: 2026-10-13 ### Memecoin vetting bot team URL: https://agenticfeats.com/c/memecoin-vetting-bot-team-grok-bot/ Grok Bot learned the author's 4-minute memecoin check from one demo, then ran it through a scout, checker, and reporter bot team. Result: Over 30 days it checked 3,140 coins, skipped 2,980, raised 160 alerts, and the author made 52 trades for +$9,320. - Agents: Grok Bot - Evidence: grade B (Numbers, a specific number stated), Author states counts and profit; a video is attached, no repo - Outcome: Completed - Job type: Research - Roles: Personal - Setup: Grok Bot copied a recorded routine across dexscreener, gmgn, rugcheck, bubblemap, and X. One bot spawned scout, checker, and reporter bots in a thread. Trades go through the separate tool jev. - Caveat: Self-reported profit with no proof. The bots can only look and write, and trades run through jev with a size limit from a small wallet. - Source: https://x.com/hrundel75/status/2108901142264487999 by @Hrundel75, 2026-10-10 - Published: 2026-10-13 ### LEGO order through Grok Bot URL: https://agenticfeats.com/c/lego-order-through-grok-bot/ Grok Bot bought a LEGO set from the LEGO website after Elon Musk asked it to order one, using a card photo shared in chat. Result: Musk reports that Grok Bot placed a real LEGO order on the LEGO website; no price, delivery, or cost savings are stated. - Agents: Grok Bot - Evidence: grade C (Claim, claim only), Author's own post on X, relayed by a news blog; no order proof shown - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Grok Bot, the agent version of Grok. The user shares a photo of a credit card in the chat and names the item to buy. - Caveat: Single self-reported order. No details on spending limits, checkout confirmation, card storage, availability, or plans that include it. - Prompt (written by ${BRAND}, not the author's exact prompt): Find the best price for [item] across online stores and order it using the card I shared in this chat. - Evidence link: https://x.com/elonmusk/status/2109028313553399947 - Source: https://teslanorth.com/2026/10/10/elon-grok-bot-shopping, 2026-10-10 - Published: 2026-10-13 ### Order logging and commission register URL: https://agenticfeats.com/c/order-logging-and-commission-register-grok-bot/ Grok Bot logged a sales order from one call, cross-checked the PI, PO and bank slip, and set up a commission register for finance. Result: It caught a one-day error in the deposit date, then also built a packing list from a two-line factory message and drafted the client email. - Agents: Grok Bot - Evidence: grade A (Shown, artifact linked), Author's own account with a photo; no repo or numbers - Outcome: Completed - Job type: Finance and Admin - Roles: Operations, Sales - Setup: Grok Bot with access to the author's order documents and Gmail. The source gives no further setup detail. - Caveat: No numbers on time saved. The author reviews the output, and only one day's work is described. - Prompt (written by ${BRAND}, not the author's exact prompt): Log the order for [client], move it to [month], and set commission to be calculated on the ship date. Check the PI against the PO and the bank slip, flag any mismatches, set up a commission register for finance, and notify me when done. - Source: https://x.com/rassailor/status/2108221352641741264 by @rassailor, 2026-10-08 - Published: 2026-10-13 ### Voice-built UI for finance desk URL: https://agenticfeats.com/c/voice-built-ui-for-finance-desk-grok-bot/ Grok Bot designed the UI/UX for the author's agentic financial desk from voice instructions, using design references it researched. Result: It produced all the pages, animated them, and made explainer videos; the author shows the output in an attached video. - Agents: Grok Bot - Evidence: grade A (Shown, artifact linked), Attached video of the designs; author states the outcome, no metrics - Outcome: Completed - Job type: Creative - Roles: Design, Founder or owner - Setup: Grok Bot on X, driven by voice, with reference designs supplied so it could research and match a style. - Caveat: No numbers or time saved stated. The post also promotes the product and speculates about the future of web design. - Prompt (written by ${BRAND}, not the author's exact prompt): Design the UI/UX for [product description]. Use these references for style: [reference links or descriptions]. Build every page, animate them, and make a short explainer video for each. - Source: https://x.com/0x3matt/status/2108797829430812912 by @0x3Matt, 2026-10-10 - Published: 2026-10-13 ### Household errands with always-on bots URL: https://agenticfeats.com/c/household-errands-with-always-on-bots-grok-bot/ Eleanor Konik used Muse and Grok Bot for library requests, dishwasher shopping, marketplace scans, party planning, grant research, and publishing a landing page. Result: Bought a new KitchenAid dishwasher within days, found a $25 armchair, caught a last-show date night, and published a landing page via Cloudflare. - Agents: Grok Bot, Other - Evidence: grade A (Shown, artifact linked), First-person blog post with specific outcomes; live site linked - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Muse (free hosted agent) plus Grok Bot (about $16/mo, bundled with Cursor Pro). Bots coordinate through a Notion task board and email. Grok Bot publishes via a Cloudflare API key. - Caveat: Muse runs a weaker model and cannot use local files. Grok Bot will not save passwords. Giant's grocery site captcha blocked the browser agent. - Evidence link: http://manuscriptions.us/ - Source: https://eleanorkonik.com/p/muse-was-my-gateway-drug-to-grok, 2026-10-12 - Published: 2026-10-13 ### Grok Bot lead generation outreach URL: https://agenticfeats.com/c/grok-bot-lead-generation-outreach/ The author built a Grok Bot agent connected to Leadsforge and Salesforge to find leads against a set ICP and draft outreach campaigns. Result: The author reports the setup added a 38% reply rate, per the post title; the visible text does not show the underlying data. - Agents: Grok Bot - Evidence: grade C (Claim, claim only), Vendor blog post; 38% reply rate claimed in title, no visible data - Outcome: Completed - Job type: Sales - Roles: Marketing, Sales - Setup: Grok Bot agent connected to Leadsforge for lead search and Salesforge for campaigns, given the company website, a specific ICP, lead-quality rules, exclusion rules, and a clear end goal. - Caveat: Author still checks the agent's work. Vendor blog, so the reply rate is self-reported. - Source: https://salesforge.ai/blog/grok-bot-lead-generation - Published: 2026-10-13 ### Parent's life admin on autopilot URL: https://agenticfeats.com/c/parent-s-life-admin-on-autopilot-grok-bot/ Grok Bot handled a parent's school-email prep, found unused gift cards, booked appointments, and listed an old wagon for sale on Facebook Marketplace. Result: It found three forgotten Gap gift cards worth $150 and booked a physical exam, a hair appointment and a massage, with the author only approving. - Agents: Grok Bot - Evidence: grade B (Numbers, a specific number stated), Author's own account on X with a photo; no repo - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Grok Bot connected to the author's email and calendar, with Zocdoc for booking. The author approves key steps through a chat-style interface. - Caveat: The author says it is not perfect. They still confirmed the last Zocdoc step and the listing price themselves. - Evidence link: https://x.com/startupstella/status/2090450591688360033 - Source: https://grokbotpulse.com/en/categories/email - Published: 2026-10-13 ### Grok Bot flight booking test URL: https://agenticfeats.com/c/grok-bot-flight-booking-test/ Grok Bot was tested as a travel agent that books flights, and the author also ran their video and social posting workflow through Grok bots. Result: The bot got close to booking flights but missed the final step. Other bots edited and uploaded the video and posted it. - Agents: Grok Bot - Evidence: grade C (Claim, claim only), Author's blog post with a video walkthrough; no direct link to the video - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: Grok Bot with separate bots for video editing, YouTube uploads, and LinkedIn posting. - Caveat: The flight booking did not complete. The bot was missing the final step. - Source: https://debbie.codes/blog/i-tested-if-grok-bot-could-book-my-flights - Published: 2026-10-13 ## Claude Cowork ### YouTube recipe to grocery cart URL: https://agenticfeats.com/c/youtube-recipe-to-grocery-cart-claude-cowork/ Claude Cowork pulled recipes from a YouTube video, matched ingredients to past-invoice brands, filled a grocery cart, and wrote a Notion meal plan. Result: It filled the grocery cart and produced a Notion recipe page, replacing a routine of 45+ minutes of watching, list-making, and cart-filling. - Agents: Claude Cowork, Claude in Chrome - Evidence: grade A (Shown, artifact linked), Screenshots and a public Notion page; author reports the time saved - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Claude Cowork with the Claude in Chrome extension, a Notion connector, a grocery site that allows bots, and a local folder of past grocery invoices. - Caveat: The Chrome extension is slow and burns usage limits fast. Items sold by weight need explicit quantity instructions. - Prompt (written by ${BRAND}, not the author's exact prompt): I want to do meal prep for the week from this video: [LINK]. Extract the recipes and ingredients, build a shopping list, and match items to products in my past invoices in [FOLDER]. Add everything to my cart at [STORE]; do not check out. Offer alternatives for unavailable items. Create a private Notion page with the weekly plan and recipes. - Evidence link: https://road-marble-ea2.notion.site/High-Protein-Meal-Prep-Plan-Week-of-Jan-27-2026-2f0d72d23a79814b8322cf26f4a55837 - Source: https://ishyshkova.substack.com/p/claude-cowork-review-real-use-cases, 2026-01-26 - Published: 2026-10-13 ### Insurance paperwork from hospital records URL: https://agenticfeats.com/c/insurance-paperwork-from-hospital-records-claude-cowork/ Claude Cowork read a folder of hospital record PDFs and filled out the author's insurance paperwork on their insurer's website. Result: The author handled complicated legal and medical insurance forms while feeding their newborn daughter, with Claude doing the filling. - Agents: Claude Cowork - Evidence: grade A (Shown, artifact linked), Author's first-person account with a video attached; no numbers - Outcome: Completed - Job type: Finance and Admin - Roles: Personal - Setup: Claude Cowork with access to a local folder of hospital records as PDFs, pointed at the insurance website. - Caveat: No time saved or accuracy check stated. Insurance forms were not independently verified in the post. - Prompt (written by ${BRAND}, not the author's exact prompt): Use the hospital records in [folder path] to fill out my insurance paperwork on [insurance website URL]. Follow the form's legal and medical wording closely, and flag anything you are unsure about before submitting. - Source: https://x.com/felixrieseberg/status/2014490352624005458 by @felixrieseberg, 2026-01-23 - Published: 2026-10-13 ### Downloads folder cleanup URL: https://agenticfeats.com/c/downloads-folder-cleanup-claude-cowork/ Claude Cowork scanned a Downloads folder of 2,200 loose files, proposed an 11-folder structure, and sorted the files after the author approved the plan. Result: The initial sort took a few minutes and left no loose files in the main directory, in about 20 minutes of active work. - Agents: Claude Cowork - Evidence: grade A (Shown, artifact linked), Author's blog post with screenshots, a sped-up video, and file counts - Outcome: Completed - Job type: Ops and IT - Roles: Personal - Setup: Claude Cowork in the Claude desktop app on a Mac with a paid plan, granted access to the Downloads folder, starting from the built-in file organization prompt. - Caveat: Needs a paid plan and macOS. The author advises backing up files first and notes prompt injection risks. - Prompt (written by ${BRAND}, not the author's exact prompt): Organize and tidy up my [folder name] folder. Group files into folders by type. Review the files in each folder, summarize what is there, and suggest how to subdivide and name them. Ask me before making changes. - Source: https://wonderingaboutai.substack.com/p/i-tried-using-claude-cowork-to-organize, 2026-01-31 - Published: 2026-10-13 ### Replacing ChatGPT with Claude Cowork URL: https://agenticfeats.com/c/replacing-chatgpt-with-claude-cowork/ Claude Cowork replaced ChatGPT for a non-technical author, who used it for slides, long-file triage, Excel files, an expense report, and a research brief. Result: The author turned 47 trip receipts into a categorized Excel expense report in 10 minutes, and built a sourced research brief in 2 minutes. - Agents: Claude Cowork - Evidence: grade B (Numbers, a specific number stated), Author's own time claims and demo videos; expense report video withheld for security - Outcome: Completed - Job type: Business - Roles: Founder or owner, Marketing - Setup: Claude desktop app, Cowork tab, Pro plan. Voice-profile and rules markdown files, the AskUserQuestion form tool, and the Gamma MCP connector for slides. - Caveat: Three blurry receipts were flagged and filled in by hand. The author says the tool is not perfect. - Prompt (written by ${BRAND}, not the author's exact prompt): We are getting sidetracked. Start over without losing this chat's context. First, generate an AskUserQuestion form so I can decide what went wrong and where to go. From my answers, generate a second form to complete [task]. - Source: https://ruben.substack.com/p/quit-chatgpt, 2026-02-04 - Published: 2026-10-13 ### German VAT filing URL: https://agenticfeats.com/c/german-vat-filing-claude-cowork/ Claude Cowork prepared quarterly accounts from invoice PDFs, then Claude in Chrome filled and submitted the German VAT forms in Elster. Result: The author's overdue Q3 2025 German VAT statements were filed in one day, with a recorded video of the submission. - Agents: Claude Cowork, Claude in Chrome - Evidence: grade A (Shown, artifact linked), Author's video of the filing, redacted by Claude and sped up 3x - Outcome: Completed - Job type: Finance and Admin - Roles: Finance, Founder or owner - Setup: Claude Cowork with a project folder of invoice and receipt PDFs, plus the Claude Chrome extension. The author logged in to Elster themselves. No custom code. - Caveat: The author logged in to the tax portal manually before handing the forms to Claude. - Prompt (written by ${BRAND}, not the author's exact prompt): Using the invoices and receipts in this folder, prepare the accounts for each quarter of [year]. Then, in the [tax portal] I have logged into, find the relevant [VAT] forms for [quarter], fill them out, and submit them. - Source: https://x.com/eastlondondev/status/2020826097890242582 by @EastlondonDev, 2026-02-09 - Published: 2026-10-13 ### Spreadsheet to KPI dashboard URL: https://agenticfeats.com/c/spreadsheet-to-kpi-dashboard-claude-cowork/ Claude Cowork read a 12-tab marketing tracker, picked key metrics, then built a color-coded KPI dashboard tab inside the live Google Sheet. Result: The author got a one-screen dashboard in about 20 minutes, work they estimate would have taken 4 to 6 hours by hand. - Agents: Claude Cowork - Evidence: grade A (Shown, artifact linked), Author states time saved; dashboard screenshot included in the post - Outcome: Completed - Job type: Business - Roles: Founder or owner, Marketing - Setup: Claude Desktop with the Cowork tab on a paid plan. The sheet was uploaded as an XLS file, then Cowork got browser access to edit the live Google Sheet. - Caveat: The time saved is the author's own estimate. Cowork needed browser access to the live sheet. - Prompt (written by ${BRAND}, not the author's exact prompt): Build me a KPI dashboard from this spreadsheet [describe data]. I need everything on one screen, color-coded with targets, weekly trends, and a clear on-track or off-track status. Before you start, show me your plan. - Evidence link: https://substackcdn.com/image/fetch/$s_!Fe-e!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F35ae28df-c8b1-4ba5-8288-65a032267c59_3276x1932.png - Source: https://michaelcrist.substack.com/p/claud-cowork-build, 2026-03-03 - Published: 2026-10-13 ### Receipt processing in Cowork URL: https://agenticfeats.com/c/receipt-processing-in-cowork-claude-cowork/ Claude Cowork processed a PDF purchasing card receipt, extracting data and producing a cover sheet for the author's TPS report, running its own Python script. Result: Cowork ran the script, fixed an error itself, and returned finished output files with no terminal or manual dependency setup from the author. - Agents: Claude Cowork - Evidence: grade C (Claim, claim only), Author's firsthand blog account; no number or output file shown - Outcome: Partial - Job type: Finance and Admin - Roles: Operations - Industry: Universities - Setup: Claude Cowork desktop research preview, pointed at a local folder of PDFs. It runs a Linux VM locally with reasoning done in Anthropic's cloud. - Caveat: The session crashed after a brief connectivity loss and needed the app's stored data cleared. Sessions are tied to one machine and not shared. - Prompt (written by ${BRAND}, not the author's exact prompt): Read the receipts in [folder], extract the key data, organize it into a spreadsheet, and produce a cover sheet for my [report name]. - Source: https://sites.duke.edu/ddmc/2026/02/24/what-claude-cowork-actually-does-and-what-it-doesnt, 2026-02-25 - Published: 2026-10-13 ### Weekly Substack ranking report URL: https://agenticfeats.com/c/weekly-substack-ranking-report-claude-in-chrome/ Claude in Chrome scrapes the top 250 Substack newsletters from Reletter each week, then Claude Cowork turns the data into a gains and losses report. Result: The author's weekly chart scraping and rank analysis now runs as two scheduled tasks every Tuesday, replacing manual copy-paste and taking a couple of minutes. - Agents: Claude Cowork, Claude in Chrome - Evidence: grade A (Shown, artifact linked), Author describes own workflow with screenshots and a video demo; no repo - Outcome: Completed - Job type: Research - Roles: Marketing - Industry: Newsletters - Setup: Claude in Chrome runs the scrape, launched from Claude Cowork. Both use scheduled tasks created by converting a working prompt, set for Tuesday at 5 PM and 5:10 PM. - Caveat: Tasks only run while the computer is awake and Claude Desktop is open. Each run uses the daily usage allowance. - Prompt (written by ${BRAND}, not the author's exact prompt): Based on the data in the folder [folder name], analyze the rankings: top 10 biggest rank gains, top 20 new entrants, top 10 biggest rank losses. Do this separately for [dataset A] and [dataset B]. - Source: https://artificialcorner.com/p/claude-automates, 2026-03-04 - Published: 2026-10-13 ### Daily AI chief of staff URL: https://agenticfeats.com/c/daily-ai-chief-of-staff-claude-cowork/ Claude Cowork runs scheduled tasks at 5 AM that triage the author's inbox, prep meetings, flag stale deals, and scan policy and literature sources. Result: The author wakes to a dashboard with Gmail drafts, prioritized actions, color-coded deals, and a personal newsletter, across about twenty scheduled tasks each week. - Agents: Claude Cowork - Evidence: grade C (Claim, claim only), First-person account by a practicing surgeon; no screenshot of the dashboard in the extract - Outcome: Completed - Job type: Business - Roles: Founder or owner, Healthcare - Industry: Medical clinics - Setup: Claude Cowork scheduled tasks on daily, weekly, and monthly cadences, with MCP connectors to Gmail, Calendar, Slack, Granola, HubSpot, and Brex. The author wrote each prompt once. - Source: https://substack.com/@techysurgeon/note/c-232846334, 2026-03-25 - Published: 2026-10-13 ### Bookkeeping and tax prep URL: https://agenticfeats.com/c/bookkeeping-and-tax-prep-claude-cowork/ Claude Cowork organized the author's tax documents, built a bookkeeping spreadsheet, and prepared a tax summary and an accountant briefing. Result: The author says it saved 6-8 hours and produced the most organized package they had given their accountant, including a discrepancy checklist. - Agents: Claude Cowork - Evidence: grade B (Numbers, a specific number stated), Author states time saved and outputs; a photo is attached but no repo or file. - Outcome: Completed - Job type: Finance and Admin - Roles: Finance, Founder or owner - Setup: Claude Cowork pointed at a local Taxes folder holding W-2s, 1099s, bank CSVs, mortgage docs, a business P&L, and last year's return. A Context.md file briefs Claude. - Caveat: Some numbers did not match and some documents were missing, so the author still had to track those down before the appointment. - Prompt (written by ${BRAND}, not the author's exact prompt): Help me set up a bookkeeping workspace for my taxes. Organize the uploaded documents into a logical folder structure. Create a Context.md by reading them and pulling out filing status, income sources, business entities, preparer info, key numbers, and open questions. Then plan what documents I still need. Build a P&L CSV with an expenses tab and an income tab. - Source: https://x.com/jjenglert/status/2041901304725774805 by @JJEnglert, 2026-04-08 - Published: 2026-10-13 ### Adyen equity research in Cowork URL: https://agenticfeats.com/c/adyen-equity-research-in-cowork-claude-cowork/ Claude Cowork pulled Adyen filings and data, built an Excel model, ran a DCF, wrote an investment memo, and set up a weekly update task. Result: A full equity research workflow on Adyen, from data pull to investment memo, was done in under 15 minutes using prompts only. - Agents: Claude Cowork - Evidence: grade A (Shown, artifact linked), Author states under 15 minutes; screenshots of folders and Excel output - Outcome: Completed - Job type: Research - Roles: Finance - Setup: Claude Cowork with access to a local research folder, yfinance for financial data, and a prompt per step. Paid data connectors or an MCP server are optional. - Caveat: Fetching earnings transcripts was slow and unreliable on the free route. The author suggests a paid data provider. - Prompt (written by ${BRAND}, not the author's exact prompt): You have access to my [TICKER] research folder. Download the last 5 annual reports and 4 recent reports into filings/, find 4 recent earnings call transcripts, and pull 5 years of prices and statements via yfinance into model/ as CSVs. Report any missing files. Do not analyse yet. - Source: https://investwithai.substack.com/p/9-steps-in-claude-cowork-to-automate, 2026-04-15 - Published: 2026-10-13 ### Dispatch phone-to-desktop tasks URL: https://agenticfeats.com/c/dispatch-phone-to-desktop-tasks-claude-cowork/ A Fortune editor used Claude Dispatch, a Cowork feature linking phone to desktop, for a month to pull files, triage email, and prep meetings. Result: From their phone, the author found lost desktop files, got key emails summarized, and ran Notion and spreadsheet tasks, cutting inbox clutter. - Agents: Claude Cowork - Evidence: grade C (Claim, claim only), First-person month-long trial in a Fortune article; no repo or demo - Outcome: Completed - Job type: Ops and IT - Roles: Operations - Setup: Claude Dispatch inside Claude Cowork, with the Claude mobile and desktop apps, a Max plan, and Slack, Notion and Chrome connected. The desktop must stay awake. - Caveat: Needs the $200 a month Max plan and an awake desktop with Claude open. Broad file-moving prompts are risky, so keep them constrained. - Prompt (written by ${BRAND}, not the author's exact prompt): For my [time] meeting, find the latest [document], summarize the last email thread, list open questions, and draft three bullets I should raise. - Source: https://fortune.com/2026/04/28/claude-dispatch-feature-capabilities-service, 2026-04-16 - Published: 2026-10-13 ### Daily intelligence briefing to Slack URL: https://agenticfeats.com/c/daily-intelligence-briefing-to-slack-claude-cowork/ Claude Cowork runs a scheduled task each morning that searches research pillars, dedupes against a Notion list, and posts a curated 10-story briefing to Slack. Result: Raghav Mehra cut morning source-checking from 45+ minutes to about 5 minutes of reading a pre-curated briefing. - Agents: Claude Cowork - Evidence: grade B (Numbers, a specific number stated), Creator's stated time saved in a roundup post; no repo - Outcome: Completed - Job type: Research - Roles: Marketing - Industry: Newsletters - Setup: Claude Cowork scheduled task at 8am CET with Notion MCP, Slack MCP, and web search. Covers 4 research pillars and a regional split. - Caveat: Only the first of the roundup's workflows is shown in the source text; the others and the dropouts are cut off. - Prompt (written by ${BRAND}, not the author's exact prompt): Every morning at [time], search [research pillars] for the latest stories, skip anything already in my [Notion reading list], pick the top [10] stories, add them to the list, and post a formatted briefing to [Slack channel]. - Source: https://buildtolaunch.substack.com/p/claude-cowork-use-cases-real-workflows, 2026-04-24 - Published: 2026-10-13 ### Cowork runs most of my work URL: https://agenticfeats.com/c/cowork-runs-most-of-my-work-claude-cowork/ Claude Cowork handles the author's inbox CRM, content audits, transcript analysis, competitor teardowns and a scheduled trend brief. Result: The author now runs about 80% of their work through Cowork and says the builds saved over 10 hours a week. - Agents: Claude Cowork - Evidence: grade A (Shown, artifact linked), Author states 10+ hours saved weekly; screenshots of reports and briefs - Outcome: Completed - Job type: Business - Roles: Founder or owner, Marketing - Setup: Claude Cowork with Gmail and Notion connected, browser use for competitor research, context files describing the business, and scheduled tasks for the daily and weekly briefs. - Caveat: Results improve a lot once Cowork has detailed context and skills set up. The 10 hours a week is a self-reported figure. - Prompt (written by ${BRAND}, not the author's exact prompt): Read my last 12 months of email in [Gmail]. Build a CRM in [Notion] covering every client, prospect and partner, with the last contact date, topic, relationship stage, and who I owe a reply. - Source: https://taylaburrell.substack.com/p/13-actually-useful-things-you-can, 2026-06-16 - Published: 2026-10-13 ### Busywork day handed to Cowork URL: https://agenticfeats.com/c/busywork-day-handed-to-cowork-claude-cowork/ Claude Cowork sorted a 200-file folder, built a weekly metrics report from several sources, and turned messy notes into a structured draft. Result: The folder was renamed and deduplicated, the report draft needed only light editing, and the notes became a structured draft with the synthesis done. - Agents: Claude Cowork - Evidence: grade C (Claim, claim only), First-person review with no screenshots or numbers for the author's own tasks - Outcome: Completed - Job type: Ops and IT - Roles: Operations - Setup: Claude Cowork in the Claude desktop app, pointed at local files and folders. The agent showed a plan first and the author approved it before work began. - Caveat: Cowork uses usage limits much faster than Chat, so heavy use points to the Max plan. Scheduled tasks are still in beta. - Prompt (written by ${BRAND}, not the author's exact prompt): Sort this folder, rename files sensibly, flag duplicates, keep originals, and tell me what is worth keeping. Folder: [path]. - Source: https://dev.to/suraj_khaitan_f893c243958/claude-cowork-review-i-handed-it-a-day-of-my-busywork-heres-what-came-back-1b92, 2026-06-20 - Published: 2026-10-13 ### Social post to lead outreach URL: https://agenticfeats.com/c/social-post-to-lead-outreach-claude-cowork/ Claude Cowork scrapes commenters on a LinkedIn post, enriches them via Apollo, scores fit against the ideal customer, and drafts DMs for the best leads. Result: The author says up to 400 signups per live event come from this, and the DMs get a 48% response rate. - Agents: Claude Cowork - Evidence: grade B (Numbers, a specific number stated), Author states 48% response rate and signup numbers; screenshots shown, no raw data - Outcome: Completed - Job type: Sales - Roles: Founder or owner, Marketing - Setup: Claude Cowork with the Apify and Apollo connectors, a folder holding company and ICP info, and the whole flow saved as a reusable skill. - Caveat: Author's own claims with no raw data shown. Apollo asks for confirmation before spending credits, and the skill must be saved manually. - Prompt (written by ${BRAND}, not the author's exact prompt): I found a LinkedIn post that speaks to my audience. Extract everyone who commented, with their comment and profile URL, using the best Apify actor. Then enrich them via Apollo, score each 0 to 5 against my ICP for [offer], and write short DMs for scores 4 and 5 that reference their comment. - Source: https://pietromontaldo.substack.com/p/find-leads-with-claude-cowork, 2026-07-08 - Published: 2026-10-13 ### Unsubscribing from unwanted newsletters URL: https://agenticfeats.com/c/unsubscribing-from-unwanted-newsletters-claude-cowork/ Claude Cowork, connected to Gmail, unsubscribed the author from unwanted email newsletters using a custom JavaScript function and can flag re-appearing senders as spam. Result: Claude unsubscribed the author from 21 newsletters in five minutes. - Agents: Claude Cowork - Evidence: grade B (Numbers, a specific number stated), Author states the count and time; article text only, no repo - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Claude Cowork on a Claude Pro plan ($20 a month) with the Gmail connector, run inside a Cowork project called Morning Gmail triage. - Caveat: Requires a paid Claude subscription. Some newsletters re-appear after unsubscribing and need to be tracked and marked as spam. - Source: https://pcworld.com/article/3185314/i-was-buried-in-email-newsletters-heres-how-claude-rescued-me.html, 2026-07-08 - Published: 2026-10-13 ### Web writing workflow with Cowork URL: https://agenticfeats.com/c/web-writing-workflow-with-cowork-claude-cowork/ Claude Cowork handles keyword research, article planning, drafting, spreadsheet tracking, and rough eye-catching images for a freelance web writer working from local folders. Result: The author now writes 40 to 50 pieces a month, up from barely one piece when they started as a beginner. - Agents: Claude Cowork - Evidence: grade B (Numbers, a specific number stated), Author states monthly volume; no repo or output samples shown - Outcome: Completed - Job type: Marketing and Content - Roles: Marketing - Industry: Freelancers - Setup: Claude Cowork in the Claude desktop app on Windows, Claude Pro plan, with Rakko Keywords linked for keyword research and project folders shared for reading and saving files. - Caveat: The author reviews every draft, limits shared folders, and checks client terms first. Higher-quality images need ChatGPT. - Prompt (written by ${BRAND}, not the author's exact prompt): Read the articles in [folder path] and summarize themes I have not written about yet in a table. Flag overlaps with past articles and suggest how to differentiate each new topic. Save the table as a file in the same folder. - Source: https://note.com/0724_1224/n/n81bad42b4fbd?hl=en, 2026-09-27 - Published: 2026-10-13 ### Job outreach email drafting URL: https://agenticfeats.com/c/job-outreach-email-drafting-claude-cowork/ Claude Cowork read an Excel of LinkedIn leads plus the author's resume and drafted a personalized Gmail email for each role. Result: Each email is tailored to the job description and resume and left ready to send, cutting hours of outreach work to minutes. - Agents: Claude Cowork - Evidence: grade A (Shown, artifact linked), Author's own claim with a screenshot; no repo or numbers. - Outcome: Completed - Job type: Sales - Roles: Personal - Setup: Claude Cowork given an Excel of LinkedIn leads and the author's resume, with Gmail used to hold the drafts. - Caveat: Drafts are left for the author to review and send manually; no reply or interview results are reported. - Prompt (written by ${BRAND}, not the author's exact prompt): Read the attached Excel of LinkedIn leads and my resume. For each role, draft a personalized email in Gmail tailored to the job description and my resume. Leave each draft unsent for my review. - Source: https://x.com/pratyush2241/status/2106348278975705117 by @Pratyush2241, 2026-10-03 - Published: 2026-10-13 ### Video to published pipeline URL: https://agenticfeats.com/c/video-to-published-pipeline-claude-cowork/ Claude Cowork transcribes a dropped-in video, proposes titles, drafts a blog post, uploads to YouTube, and schedules social posts while the author reviews each step. Result: The author ran the full pipeline several times; it still takes over the one-hour daily goal, down from a typical 3-4 hours by hand. - Agents: Claude Cowork - Evidence: grade C (Claim, claim only), Author's own account of the workflow; no demo or numbers beyond rough time estimates - Outcome: Partial - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Youtubers - Setup: Claude Cowork project with an inbox folder, local Whisper transcription, a style guide, GitHub pull requests with Netlify previews, Figma thumbnail template, YouTube upload, and a scheduled task for Bluesky. - Caveat: Each run still needs small system tweaks and takes over an hour. The author says the daily-publishing hypothesis is not yet proven. - Source: https://seancdavis.com/posts/how-i-built-a-video-to-published-pipeline-with-claude-cowork - Published: 2026-10-13 ## Codex ### Parallel feature shipping in Codex URL: https://agenticfeats.com/c/parallel-feature-shipping-in-codex/ Keith Rumjahn used the OpenAI Codex desktop app with git worktrees and parallel threads to build four features for their Mac app at once. Result: Four features finished in about 2 hours instead of 3-5 days each, and the daily view took 40 minutes. - Agents: Codex - Evidence: grade A (Shown, artifact linked), Author states time saved and feature list; video linked, no repo - Outcome: Completed - Job type: Coding - Roles: Developer, Founder or owner - Industry: Indie hackers - Setup: Codex desktop app with four threads in Work Tree mode, heavy refactoring sent to Codex Cloud, plus saved Skills for Linear tickets and YouTube analytics. Requires ChatGPT Plus. - Caveat: Git worktrees have a learning curve, and Codex only works with OpenAI models. Their YouTube view gains are only partly credited to the skill. - Evidence link: https://www.youtube.com/watch?v=EwVs3O2Zm6I - Source: https://rumjahn.substack.com/p/the-complete-guide-to-openai-codex, 2026-02-06 - Published: 2026-10-13 ### Codex agent as company employee URL: https://agenticfeats.com/c/codex-agent-as-company-employee/ The author built an autonomous agent from Codex, ChatGPT Plus, systemd, and UNIX email and user accounts, and uses it like an employee. Result: The agent handles all kinds of tasks for their company, and the author reports no memory problems so far. - Agents: Codex - Evidence: grade C (Claim, claim only), Author's own claim in an HN comment; no repo or demo - Outcome: Completed - Job type: Business - Roles: Founder or owner - Setup: Codex with ChatGPT Plus, run under systemd, using normal UNIX email and user accounts. Memory is one long rolling context window, a diary, and a markdown wiki folder. - Caveat: Early days. The author expects problems as demands grow and may add specialist bots. - Source: https://news.ycombinator.com/item?id=47730086 by mike_hearn, 2026-04-11 - Published: 2026-10-13 ### Amazon subscription refund URL: https://agenticfeats.com/c/amazon-subscription-refund-codex/ Codex running GPT-5.5 cancelled an Amazon subscription, opened the live chat with an associate, and argued for an immediate refund. Result: The associate confirmed the refund and the author got $15 back within about five minutes, while the author watched it happen. - Agents: Codex - Evidence: grade A (Shown, artifact linked), Author states the $15 refund; a video is attached - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Codex with GPT-5.5 navigating Amazon account settings, the help page, and the live chat with an associate. - Caveat: The refund was small at $15. The author only describes one run, and the source does not show the chat transcript. - Prompt (written by ${BRAND}, not the author's exact prompt): Cancel my [subscription] on [service] and get a refund for the current billing period. If asked, open the live chat with an associate and explain that I want it stopped now and refunded. - Source: https://x.com/chrisgpt/status/2049341524958151000 by @ChrisGPT, 2026-04-29 - Published: 2026-10-13 ### Codex email audit and trip site URL: https://agenticfeats.com/c/codex-email-audit-and-trip-site/ Codex, driven by two authors, audited a month of missed emails, built a vacation planning site from a phone, and shopped for a dishwasher. Result: Codex returned a complete, classified list of missed email threads where Claude missed many; it also deployed a vacation site researched from hotel availability. - Agents: Codex - Evidence: grade A (Shown, artifact linked), Live podcast and Peloton dashboard links; email comparison is author's claim - Outcome: Completed - Job type: Personal and Family - Roles: Founder or owner, Personal - Setup: Codex app with GPT-5.5, Gmail, Google Calendar, Drive and Slack plugins, built-in browser, auto-review permissions mode, and remote control from a phone. - Caveat: Fewer native connectors than Claude, occasional frustrations, and the authors still prefer Claude for design and writing work. - Prompt (written by ${BRAND}, not the author's exact prompt): Review my email from the past [time period] and list every thread I have not responded to. Classify each one by [categories] and flag the most urgent. - Evidence link: https://clairem0.github.io/technically-curious-peloton-dashboard/dashboard/dashboard.html - Source: https://technicallycurious.substack.com/p/codex-is-excellent, 2026-05-27 - Published: 2026-10-13 ### Client notes to Belmond shortlist URL: https://agenticfeats.com/c/client-notes-to-belmond-shortlist-codex/ Codex Desktop read a client notes Word doc, checked Belmond's live booking site, and built a three-option comparison matching the client's past trip preferences. Result: Codex returned a ranked comparison of three Belmond properties with availability and nightly rate ranges for a sample window, plus a more detailed Word doc. - Agents: Codex - Evidence: grade B (Numbers, a specific number stated), Author pastes the Codex output; no repo or screenshot linked - Outcome: Completed - Job type: Research - Roles: Founder or owner, Sales - Industry: Travel agencies - Setup: Codex Desktop app on Mac or Windows, pointed at a local client folder, using its web browsing. Works with a free ChatGPT account or Plus. - Caveat: Not ready to send to a client as is. Exact pricing needs a second pass once traveler count is known. - Prompt (written by ${BRAND}, not the author's exact prompt): Look through this client folder for their preferences and past trips. Then check the current [brand] website for availability and new properties. Create a comparison of three options and destinations that match what they liked last time. - Source: https://worldviatravelnetwork.com/blog/ai-tip-use-codex-desktop-to-connect-your-files-chatgpt-and-the-web, 2026-06-18 - Published: 2026-10-13 ### Research workflow automation with Codex URL: https://agenticfeats.com/c/research-workflow-automation-with-codex/ Codex runs a weekly arXiv search, triages new papers, feeds a Zotero library, and turns hypotheses into cheap experiment tests for an AI safety researcher. Result: The author cut time on literature triage and experiment setup, and says related-work sections now need less citation double-checking. No numbers are given. - Agents: Codex - Evidence: grade C (Claim, claim only), First-person account with screenshots and a GIF; no figures stated - Outcome: Partial - Job type: Research - Roles: Developer - Setup: Codex app with weekly automations, a CSV of seen papers, Zotero with a vector database, subagent threads, Bitwarden Secrets Manager for scoped credentials, and a hook that blocks destructive commands. - Caveat: A temporary HOME variable made Codex run cleanup against the author's real WSL home folder, because the destructive-command hook was not enabled there. - Prompt (written by ${BRAND}, not the author's exact prompt): Here is my hypothesis: [hypothesis]. Using the repo context and notes in [paths], propose cheap tests, confounders, success criteria and scrap criteria. Return paths, commands, logs and examples to inspect, plus the next cheapest check. - Source: https://aisagroup.substack.com/p/how-i-use-codex-to-automate-parts, 2026-07-07 - Published: 2026-10-13 ### Comms tasks with Codex URL: https://agenticfeats.com/c/comms-tasks-with-codex/ The author used OpenAI's Codex to build a PR coverage dashboard, sort a messy work folder, and filter a recipe site for picky kids. Result: Codex sorted 240 dinner recipes down to 62 family-fit picks, rated them, and produced an index, a spreadsheet, and a Google Drive folder. - Agents: Codex - Evidence: grade B (Numbers, a specific number stated), Author's own account with counts (240 to 62 recipes); no repo or output shown - Outcome: Completed - Job type: Personal and Family - Roles: Marketing, Personal - Setup: Codex desktop app with a ChatGPT login, GPT-5.5 model, access to local files, and a Google Drive connection. - Caveat: The author says they are early in learning the tool and that its work should not run without supervision. - Prompt (written by ${BRAND}, not the author's exact prompt): Visit [recipe website] and pull only the dinner recipes. Filter for ones that suit [picky eaters]. Group and rate them as Best Bets, Solid Weeknight Options, and Maybe with Modifications. Build an index, a sortable spreadsheet, and a Google Drive folder. - Source: https://youbetabelieveit.substack.com/p/oh-holy-codex-the-ai-coworker-has, 2026-06-23 - Published: 2026-10-13 ### Slack issue triage to Codex fix URL: https://agenticfeats.com/c/slack-issue-triage-to-codex-fix/ An internal ops agent triaged a Slack issue, passed it to a Codex agent, and then verified the fix back in Slack. Result: The Codex agent resolved the issue in 40 minutes, and the ops agent confirmed the fix in Slack. - Agents: Codex, Custom agent - Evidence: grade B (Numbers, a specific number stated), Author states the 40 minute fix; no repo or screenshot - Outcome: Completed - Job type: Ops and IT - Roles: Operations - Setup: A purpose-built internal ops agent monitors Slack and hands issues to a Codex agent, then checks the outcome. - Caveat: The source does not say what the issue was or how often this works; it is a single reported case. - Source: https://x.com/wesleymagness/status/2080314934601433570 by @wesleymagness, 2026-07-23 - Published: 2026-10-13 ### Self-hosted agentic software factory URL: https://agenticfeats.com/c/self-hosted-agentic-software-factory-hermes-agent/ The author built a sandboxed home server stack where Hermes with Codex takes one prompt and runs the whole software lifecycle, from repo to deployment. Result: From one prompt it created a repo, wrote the app and tests, got CI green, provisioned Postgres and deployed behind HTTPS with no further messages. - Agents: Codex, Hermes Agent - Evidence: grade A (Shown, artifact linked), Blog post with demo video link and setup repos - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Hermes (Codex inference) on an isolated i7 box with Coolify, Forgejo and runners, self-hosted Firecrawl, Pi-hole, Tailscale, Telegram, Porkbun DNS-01 and Let's Encrypt certs. - Caveat: Inference, Tailscale, Telegram, DNS and ACME still leave the box. Setup is not a full how-to and needs a £20 Codex subscription. - Evidence link: https://inventronix.lon1.cdn.digitaloceanspaces.com/demo_web.mp4 - Source: https://blog.jakesaunders.dev/building-an-almost-fully-self-hosted-sandboxed-agentic-software-factory, 2026-08-21 - Published: 2026-10-13 ### Find why a process dies URL: https://agenticfeats.com/c/find-why-a-process-dies-codex/ Codex CLI, run on the author's Mac, connected to their Ubuntu server over SSH and checked kernel logs to find why Claude Code kept dying. Result: Codex traced the kills to the Linux OOM killer on a 2 GB droplet with no swap, and listed fixes without changing anything. - Agents: Codex - Evidence: grade C (Claim, claim only), Blog post with the session transcript and log excerpts; no repo - Outcome: Completed - Job type: Ops and IT - Roles: Founder or owner, Operations - Setup: Codex CLI on a Mac with SSH access to a DigitalOcean Ubuntu 24.04 droplet using a key. It read journalctl kernel logs, free, ps and ulimit output. - Caveat: The author resized the droplet instead of adding swap, which was the cheaper fix, and that caused later problems. - Prompt (written by ${BRAND}, not the author's exact prompt): Connect to my server with [ssh command]. When I start [process] on it, it is killed within seconds. Find out why, show the evidence, and do not change anything yet. - Source: https://flaviocopes.com/server-process-killed, 2026-09-15 - Published: 2026-10-13 ### Substack stats report in Codex URL: https://agenticfeats.com/c/substack-stats-report-in-codex/ Codex used Computer Use to open the author's Substack dashboard and pull figures for every published post into a newsletter report. Result: The author uses Codex daily for reports, editable designs, video editing and this newsletter, which still needed plenty of corrections. - Agents: Codex - Evidence: grade A (Shown, artifact linked), Guide with screenshots from the author's own Codex recordings; no numbers - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Newsletters - Setup: Codex desktop app with a project folder, GPT-6 Astra model, and the Computer Use plugin (server and skill enabled) for browser clicks. - Caveat: Newsletter output still took plenty of corrections. Full access permissions should be limited to what the job needs and checked afterward. - Prompt (written by ${BRAND}, not the author's exact prompt): Open my [Substack dashboard], pull the figures for every published post, and build a report in this project folder. Check what saved before finishing. - Source: https://charliehills.substack.com/p/codex-beginner-advanced, 2026-09-20 - Published: 2026-10-13 ### Faster Rust ML code via benchmarks URL: https://agenticfeats.com/c/faster-rust-ml-code-via-benchmarks-codex/ Coding agents wrote Rust ML algorithms like UMAP from scratch, then iteratively optimized them against criterion benchmarks. Result: Each model pass added about 1.5x to 2.0x speedup, reaching roughly 7.5x to 32x over baseline; one implementation beat xgboost on speed. - Agents: Codex, Other - Evidence: grade A (Shown, artifact linked), Blog post with prompts, criterion output and benchmark charts; some linked repos and gist - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Agents in Zed Agent with long Markdown prompts; Rust criterion benchmarks, faer, simsimd, rayon, PyO3 bindings, no unsafe code, quality checked against umap-learn. - Caveat: Optimizing against benchmarks can overfit to them; vague goals failed, and speed gains needed a quality check against a reference. - Prompt (written by ${BRAND}, not the author's exact prompt): Run the CPU benchmarks first to set a True Performance Baseline. Then optimize the library code so all benchmarks run at least [1.2x] faster than baseline, without editing the benchmarks or adding unsafe code. Repeat until results stop improving, reporting relative speedup after each iteration. - Evidence link: https://gist.github.com/minimaxir/86de3cc8f628079d8337e70924b3411d - Source: https://minimaxir.com/2026/09/agentic-iteration, 2026-09-21 - Published: 2026-10-13 ### Ableton Live MCP for Codex URL: https://agenticfeats.com/c/ableton-live-mcp-for-codex/ The author built an MCP server so Codex can control Ableton Live, including running Python inside Live. Result: Codex built a full EDM track and a piano duet from scratch in minutes in Ableton, shown in two demo videos with the prompts listed. - Agents: Codex - Evidence: grade A (Shown, artifact linked), Repo README with two demo videos, prompts, and a spectrogram image - Outcome: Completed - Job type: Creative - Roles: Personal - Industry: Musicians - Setup: Codex CLI with a custom Ableton Live MCP server, an Agent Audio Tap Max for Live device, and Ableton's Object model. Tested on Live Suite 12.3.8, macOS. - Caveat: Only tested on macOS with Live 12.3.8. The MCP edits the set directly and could corrupt it, so back up first. - Prompt (written by ${BRAND}, not the author's exact prompt): In Ableton, make [song concept] using MIDI and Ableton devices for every instrument except [audio parts]. Then refine: [follow-up change, such as more dynamics or sidechain]. - Evidence link: https://www.youtube.com/watch?v=8dRRrIY7NI0 - Source: https://github.com/bschoepke/ableton-live-mcp - Published: 2026-10-13 ### First personal dinner-planning app URL: https://agenticfeats.com/c/first-personal-dinner-planning-app-codex/ A non-programmer used Codex to build a local web app that suggests dinners from pantry, garden, dietary rules, appliances and weather. Result: Built a working What's for Dinner app with pantry, garden, recipe and restaurant pages plus a suggestions engine, in under a week. - Agents: Codex - Evidence: grade A (Shown, artifact linked), Blog post with screenshots and a detailed build description; no public repo - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Codex working in a local project folder. ChatGPT drafted the starting prompt. Stack: Python, FastAPI, SQLite, uvicorn, Jinja2. Runs in a browser on the author's Mac. - Caveat: Personal use only, runs on one laptop. The author had to approve Codex's terminal and file operations on macOS. - Prompt (written by ${BRAND}, not the author's exact prompt): You are a careful senior engineer building for a non-coder. Project: [app name], a local macOS app, no cloud. Use [backend], SQLite, and simple HTML. Build in small milestones, and after each say how to run it. Milestone 1: CRUD screens for [list of data types]. Write a README with run steps. - Source: https://beginningwithi.com/2026/03/03/retirement-hobby-built-my-first-app-in-under-a-week - Published: 2026-10-13 ### Personal site redesign with Codex URL: https://agenticfeats.com/c/personal-site-redesign-with-codex/ Codex CLI redesigned the author's personal site after reading a design prompt copied from Anthropic's frontend-design plugin into a local markdown file. Result: The author shipped a redesigned personal site, nearly right in one shot, after a few more iterations. - Agents: Codex - Evidence: grade A (Shown, artifact linked), Blog post with before and after screenshots and session screenshot; no repo - Outcome: Completed - Job type: Coding - Roles: Developer, Founder or owner - Setup: Codex CLI, with the long-form frontend-design prompt from Anthropic's Claude Code plugin saved as a markdown file in the project and read first. - Caveat: A few rounds of iteration were still needed, and the author's own taste guided the direction. - Prompt (written by ${BRAND}, not the author's exact prompt): Read [design-spec.md] and treat it as the design specification for my [personal site]. Then build the UI for [page or section], following its guidance on typography, spacing, and hierarchy. - Source: https://wilbertliu.com/blog/how-i-designed-my-personal-site-using-codex-cli - Published: 2026-10-13 ### Stolen package refund via Computer Use URL: https://agenticfeats.com/c/stolen-package-refund-via-computer-use-codex/ Codex with Computer Use waited in Amazon's support queue, polling on a schedule, then pursued a refund for a stolen package. Result: The author showered while Codex waited in the queue and negotiated. The refund was done when they returned. - Agents: Codex - Evidence: grade C (Claim, claim only), Author's first-person account; no screenshot or log shown - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Codex with Computer Use plus a Goal (get the refund) and a thread automation waking the thread every five minutes, then every minute once an agent joined. - Caveat: The author stays present for payments, credentials, account changes, and security settings, and lets the agent only research and draft there. - Prompt (written by ${BRAND}, not the author's exact prompt): Use Computer Use to contact [company] support about [problem]. Check the queue every five minutes. Once a person joins, check every minute and do your best to get me [desired outcome]. - Source: https://jxnl.co/writing/2026/06/16/three-ways-codex-can-use-a-computer - Published: 2026-10-13 ### AI chat app on Nokia 110 URL: https://agenticfeats.com/c/ai-chat-app-on-nokia-110-codex/ Codex helped the author reverse-engineer a Nokia 110 4G's firmware and build a native AI chat app with a custom agent. Result: A working prototype runs on the handset over mobile data: typed chat via DeepSeek, battery check, torch, calls and alarms, tested on the phone. - Agents: Codex, Custom agent - Evidence: grade A (Shown, artifact linked), Demo video and README; source code not published - Outcome: Partial - Job type: Coding - Roles: Developer - Setup: OpenAI Codex, Calculator app entry point modified into a chat app, DeepSeek chat API with tool calls, custom lightweight agent to fit 48MB memory. - Caveat: Code is loaded into RAM from a computer and must be reloaded after restart or power-off. Source code is not published. - Evidence link: https://www.youtube.com/watch?v=i5Ce53QkMkU - Source: https://github.com/anupray95/AI-Agent-on-a-NOKIA - Published: 2026-10-13 ## OpenClaw ### Home Assistant AI agent with persona URL: https://agenticfeats.com/c/home-assistant-ai-agent-with-persona-openclaw/ OpenClaw, running in a Proxmox LXC beside Home Assistant via ha-mcp, became an always-on agent with persistent markdown memory and a raccoon persona named Claudette. Result: Claudette controls the lights and speaks through Google Home speakers in an ElevenLabs voice, shown in a live demo. - Agents: OpenClaw - Evidence: grade A (Shown, artifact linked), Blog post with architecture diagrams and chat screenshots; no public repo of the setup - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: OpenClaw in an Ubuntu LXC (2GB RAM, 2 cores) connected to Home Assistant through ha-mcp, with Telegram, Gmail, ElevenLabs voice, and markdown memory files synced nightly to GitHub. - Caveat: The visible text covers only a few demos; the security and cost section is cut off, and the setup needs Claude API access. - Source: https://dan-malone.com/blog/openclaw-home-assistant, 2026-01-30 - Published: 2026-10-13 ### Agent-built research knowledge bases URL: https://agenticfeats.com/c/agent-built-research-knowledge-bases-openclaw/ OpenClaw learned to use a YouTube summariser, a paper extraction tool, and QMD, then built searchable Markdown knowledge bases for videos and research papers. Result: The author now has two dense, searchable knowledge bases, one from YouTube videos and one from research papers, which they say speeds up understanding. - Agents: OpenClaw - Evidence: grade A (Shown, artifact linked), Screenshots of agent Q&A; author's own report, no metrics - Outcome: Completed - Job type: Research - Roles: Personal - Setup: OpenClaw agent on a dedicated Mac mini. Author installed youtube-summariser via pip and Tobi's QMD via bun, plus their own paper-fluff-cutter tool. The agent learned each tool. - Caveat: No numbers on size or time saved; the benefit is the author's own impression. - Prompt (written by ${BRAND}, not the author's exact prompt): Learn how to use [summariser tool] and [QMD], then use both to build your own knowledge base from [source type]. Save outputs as Markdown so they can be indexed and searched. - Source: https://notesfromzero.substack.com/p/building-knowledge-bases-for-my-openclaw, 2026-02-05 - Published: 2026-10-13 ### OpenClaw personal agent setup URL: https://agenticfeats.com/c/openclaw-personal-agent-setup/ The author runs OpenClaw on an old MacBook Air, talks to it through Slack, and set up a daily morning brief and a task dashboard. Result: A working personal agent that sends an 8am brief and updates a vibe-coded Kanban dashboard in real time as it works. - Agents: OpenClaw - Evidence: grade A (Shown, artifact linked), Author's own account with dashboard and brief screenshots; no repo - Outcome: Partial - Job type: Setup and Meta - Roles: Founder or owner, Personal - Setup: OpenClaw on a wiped old MacBook Air, chat via Slack and Telegram, connected to OpenAI and Gemini keys, Notion, Brave search, Nano Banana. Dashboard built with Converge and Convex. - Caveat: Insecure by default, since it holds API keys and can reach everything on its host machine. The subreddit scan in the morning brief broke. - Prompt (written by ${BRAND}, not the author's exact prompt): Send me a morning brief every morning at [time] my time. Include local weather, trending videos and stories on [interests], my tasks for today from [todo list], and tasks you can take off my plate. - Source: https://prompthub.substack.com/p/how-im-using-openclaw, 2026-02-17 - Published: 2026-10-13 ### VC analyst agent and CRM build URL: https://agenticfeats.com/c/vc-analyst-agent-and-crm-build-openclaw/ OpenClaw, set up as a VC analyst agent named Pao, audited its own server security and coded a basic CRM. Result: Pao fixed firewall, root SSH and Docker proxy issues for $0.30, and a basic internal CRM was built for about $25 in one weekend. - Agents: OpenClaw - Evidence: grade A (Shown, artifact linked), Screenshot of V1 CRM; author states costs and audit result - Outcome: Partial - Job type: Business - Roles: Founder or owner, Operations - Industry: Financial advisors - Setup: OpenClaw on a $6.99/month Hostinger VPS, Claude API, OpenRouter routing, Telegram bot for mobile access, Supabase, Vercel and GitHub. Setup took about 45 minutes. - Caveat: Claude hit usage limits and stalled deployments. Kimi orchestrated poorly, and it is hard to verify sub-agent work. - Evidence link: https://substackcdn.com/image/fetch/$s_!CtqD!,w_2400,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6710ca3f-32a3-4e7b-82d0-dee2eb3f6a83_1870x797.png - Source: https://shriftman.substack.com/p/my-first-experience-with-openclaw, 2026-02-09 - Published: 2026-10-13 ### OpenClaw buys from Amazon URL: https://agenticfeats.com/c/openclaw-buys-from-amazon/ OpenClaw agent used the buy-anything skill and Rye's checkout API to find, check out, and order products for Ryan Eade. Result: Ryan completed real purchases delivered to his door, replaced his Alexa buying flow, and passed the Amazon challenge on a live podcast demo. - Agents: OpenClaw - Evidence: grade A (Shown, artifact linked), Vendor blog with user quote and linked podcast demo video - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: OpenClaw with the buy-anything skill, Rye Universal Checkout API key, Stripe tokenized card, saved shipping info in agent memory, and a set spending limit. - Caveat: Written by the vendor, Rye. Live demo covered Amazon; Shopify support is new in v2. Agents with purchasing power carry risk. - Evidence link: https://www.youtube.com/watch?v=Ta7yT1Z1GkE - Source: https://rye.com/blog/openclaw-buy-anything-skill-agentic-checkout, 2026-03-06 - Published: 2026-10-13 ### Notion control plane for agent fleet URL: https://agenticfeats.com/c/notion-control-plane-for-agent-fleet-openclaw/ The author built AgentOps, a Notion control plane that tracks, queues, logs, and alerts on their 18 OpenClaw agents. Result: Notion now tracks 18 agents and 78 logged runs; separately, an unsubscribe agent cleared 47 vendor mailing lists in two weeks. - Agents: OpenClaw - Evidence: grade A (Shown, artifact linked), Blog with dashboard screenshots and a live newsletter site link - Outcome: Completed - Job type: Ops and IT - Roles: Developer - Setup: OpenClaw agents on Lightsail or a Raspberry Pi, connected to four Notion databases (Agent Registry, Task Queue, Run Log, Alerts) through Notion MCP. Status changes in Notion control dispatching. - Caveat: The serverless AgentCore variant is described as experimental and not production ready. - Evidence link: https://theagenticengineer.waltsoft.net/ - Source: https://dev.to/aws-heroes/i-turned-notion-into-a-control-plane-for-my-18-openclaw-ai-agents-5624, 2026-03-07 - Published: 2026-10-13 ### Car purchase negotiation URL: https://agenticfeats.com/c/car-purchase-negotiation-openclaw/ OpenClaw researched Palisade prices, contacted nearby dealers, and played their quotes against each other by email for AJ Stuyvenberg. Result: The agent negotiated a $4,200 dealer discount, bringing the car to about $56k, and the buyer never visited a showroom. - Agents: OpenClaw - Evidence: grade B (Numbers, a specific number stated), Author's own account on a personal blog, retold by a third-party post; no artifact - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: OpenClaw on an always-on machine with Gmail and browser access. A cron job checked email every few minutes and sent each dealer the lowest competing quote. - Caveat: The agent once sent a message to the wrong dealer, and the buyer handled the credit application and pickup himself. - Prompt (written by ${BRAND}, not the author's exact prompt): Research typical prices for [car model and trim] in [region]. Find dealer inventory within [distance] of [city] matching [spec]. Contact each dealer for a quote, then send each one the lowest competing offer and ask them to beat it. - Evidence link: https://aaronstuyvenberg.com/posts/clawd-bought-a-car - Source: https://cloudship.co.uk/blog/ai-agent-bought-a-car, 2026-03-04 - Published: 2026-10-13 ### Real estate agent lead scraping URL: https://agenticfeats.com/c/real-estate-agent-lead-scraping-openclaw/ OpenClaw scraped Zillow, Redfin and Realtor.com for agents listing 1965 to 2000 homes in the author's service area. Result: A CSV of agent leads with brokerage, phone, email, active listings, addresses and year built landed in Google Drive, and outreach had started. - Agents: OpenClaw - Evidence: grade C (Claim, claim only), Author's own account; no screenshot, lead count, or repo - Outcome: Completed - Job type: Sales - Roles: Founder or owner, Sales - Industry: Roofing - Setup: OpenClaw agent named Cody, driven by prompts over Discord from a phone. Output saved to Google Drive. - Caveat: No lead count, reply rate, or booked jobs reported. Outreach had only just begun when the post was written. - Prompt (written by ${BRAND}, not the author's exact prompt): Scrape Zillow, Redfin and Realtor.com for real estate agents with active listings of homes built between [start year] and [end year] in [service area]. Collect brokerage, phone, email, listing addresses and year built. Save as a CSV to Google Drive, then start outreach. - Source: https://x.com/pricefoulger/status/2033271956829806891 by @pricefoulger, 2026-03-15 - Published: 2026-10-13 ### Email triage and drafting agent URL: https://agenticfeats.com/c/email-triage-and-drafting-agent-openclaw/ An agent called Donna triaged three email accounts for 30 days, flagged urgent mail on Telegram, and drafted routine replies for approval. Result: About 85% of triage was correct, no urgent email was missed in 30 days, and routine inbox processing dropped from 10 minutes to 2. - Agents: OpenClaw - Evidence: grade B (Numbers, a specific number stated), Author's own 30-day report with percentages; no repo or screenshots - Outcome: Partial - Job type: Business - Roles: Founder or owner, Operations - Setup: Donna runs on an OpenClaw gateway on a Mac Mini with IMAP and Microsoft Graph access, a persistent memory file, 30-minute checks, and Telegram alerts. Never sends without approval. - Caveat: Tone matching, context loss across sessions, CC choices, and relationship-sensitive replies failed, so the scope was cut back to routine emails. - Prompt (written by ${BRAND}, not the author's exact prompt): Check [email accounts] every 30 minutes in business hours. Sort mail into urgent, important, informational, and noise. Message me on [channel] for urgent items. Draft replies to routine mail only, and never send without my approval. - Source: https://dev.to/jarveyspecter/i-let-an-ai-agent-manage-my-emails-for-30-days-heres-what-broke-4n12, 2026-03-08 - Published: 2026-10-13 ### Self-updating social media agent URL: https://agenticfeats.com/c/self-updating-social-media-agent-openclaw/ OpenClaw ran on a schedule across 18 cron jobs, analyzed two weeks of engagement data on X and Bluesky, and rewrote its own strategy file. Result: The agent told the author to stop posting on X and focus on Bluesky. The author followed the advice without having asked for it. - Agents: OpenClaw - Evidence: grade C (Claim, claim only), Author's own account from a paywalled live demo; no repo or numbers - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Setup: OpenClaw with 18 cron jobs, Telegram as the interface, skills managed in GitHub, and MCP connections to outside tools such as Google Search Console. - Caveat: The full walkthrough is behind a paywall, and no engagement numbers or results of the switch are given. - Source: https://buildtolaunch.substack.com/p/openclaw-autonomous-agent-live-demo, 2026-03-16 - Published: 2026-10-13 ### Personal memory from data exports URL: https://agenticfeats.com/c/personal-memory-from-data-exports-openclaw/ OpenClaw, fed ChatGPT, Google and Instagram history via context-use, books restaurants and sends daily recommendations. Result: The agent found and booked a restaurant the author visited, and recommended a book that the author read and liked before a trip to Kerala. - Agents: OpenClaw - Evidence: grade C (Claim, claim only), Author's own account; no repo or screenshots for the result - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: OpenClaw with the context-use tool, loaded with exported ChatGPT, Google and Instagram data. The author edits and approves a daily memory diff. - Caveat: The author works on the context-use team. The memory updates need manual review and approval each day. - Source: https://news.ycombinator.com/item?id=47476742 by PrincessEe, 2026-03-22 - Published: 2026-10-13 ### Complex personal tax return URL: https://agenticfeats.com/c/complex-personal-tax-return-openclaw/ OpenClaw read years of tax records and entity documents, built a summary, then guided the author through filing a complex return in TurboTax. Result: Filed a 137-page federal return for $129 in software and $23.90 in tokens versus an $8,000 CPA bill, in about 6 hours. - Agents: OpenClaw - Evidence: grade B (Numbers, a specific number stated), Author's detailed account with costs and time; no repo or screenshots - Outcome: Completed - Job type: Finance and Admin - Roles: Personal - Setup: OpenClaw with GPT 5.4 reading PDFs and CSVs of prior returns, K-1s, agreements and bank statements; the author entered everything into TurboTax DIY by hand. - Caveat: Browser automation of TurboTax was too slow, so the author typed while OpenClaw dictated. K-1 income was estimated, so an amended return may be needed. - Prompt (written by ${BRAND}, not the author's exact prompt): Read everything in [folder] in chronological order and work out my [year] tax situation. List the forms needed, key numbers, how past returns treated each item, and open questions in summary.md. - Source: https://x.com/dvassallo/status/2039450517143974105 by @dvassallo, 2026-04-01 - Published: 2026-10-13 ### CI failure triage agents URL: https://agenticfeats.com/c/ci-failure-triage-agents-openclaw/ OpenClaw agents watched GitHub Actions failures, classified them, and posted analysis and issues, then grew into test generation, coverage scanning, and self-healing locators. Result: Time to diagnose a CI failure fell from 45 minutes to 6, and test flakiness dropped from 8% to 0.8%. - Agents: OpenClaw - Evidence: grade B (Numbers, a specific number stated), Author's own blog post with numbers; no repo or demo of the agents - Outcome: Completed - Job type: Coding - Roles: Developer, Operations - Setup: Self-hosted OpenClaw agents with access to GitHub Actions, GitHub issues, Slack, and Playwright tests. Confidence thresholds gate auto-merges. - Caveat: A self-healing agent once changed an assertion instead of the locator. API costs neared $4,000 a month before prompt trimming cut them to $400. - Prompt (written by ${BRAND}, not the author's exact prompt): Read the failed CI run logs for [repo]. Classify the failure as flaky, real regression, or environment issue. Post the analysis to [Slack channel] and open a GitHub issue with context and a confidence score. - Source: https://suneetmalhotra.com/blog/openclaw-production-ai-agent-runtime-2026, 2026-04-08 - Published: 2026-10-13 ### Nine OpenClaw agents URL: https://agenticfeats.com/c/nine-openclaw-agents/ Claire Vo built nine OpenClaw agents covering sales, business operations, scheduling, family logistics, and kids' education, after starting with a single executive-assistant agent. Result: One agent now does the CRM management and customer email drafting that cost about 10 paid hours a week last year. - Agents: OpenClaw - Evidence: grade B (Numbers, a specific number stated), Founder's own account on a podcast, reported by Business Insider; no repo - Outcome: Completed - Job type: Business - Roles: Founder or owner, Personal - Setup: OpenClaw agents run across several computers. Access grows through a progressive trust process: calendar first, then email visibility, then drafting and sending, then more autonomous tasks. - Caveat: Her first OpenClaw attempt wiped her family calendar. She also flags risks of file deletion and the agent knowing where her kids go to school. - Source: https://aol.com/articles/startup-founder-explains-why-she-054950405.html, 2026-04-01 - Published: 2026-10-13 ### Freelance ops on autopilot URL: https://agenticfeats.com/c/freelance-ops-on-autopilot-openclaw/ OpenClaw sends a morning PR and Slack briefing, drafts Friday client status emails, and queues follow-up drafts for approval on Telegram. Result: The author says weekly admin dropped from about 9 hours to a short morning review on Telegram, with setup taking 47 minutes. - Agents: OpenClaw - Evidence: grade B (Numbers, a specific number stated), Author's own write-up with skill files and sample output; no repo or demo - Outcome: Completed - Job type: Business - Roles: Developer, Founder or owner - Industry: Freelancers - Setup: OpenClaw gateway on a Mac Mini with Claude via the Anthropic API, three skill files, one Lobster workflow, and the gh, Slack, Linear and Gmail CLIs. Approvals happen in Telegram. - Caveat: Time saved is self-reported with no before and after measurement. The follow-up detection skill is described but its file is not shown. - Prompt (written by ${BRAND}, not the author's exact prompt): Every Friday at 4PM, read this week's merged PRs and closed tickets for [client], then draft a short plain-prose status email: what shipped, what is in progress, and any decision needed from the client. Save it to [drafts folder] and send me the draft to approve. - Source: https://dev.to/farooqshabbir/i-built-a-247-freelance-operations-agent-with-openclaw-it-runs-my-business-while-i-sleep-4lde, 2026-04-23 - Published: 2026-10-13 ### Office grocery ordering via ChatGPT URL: https://agenticfeats.com/c/office-grocery-ordering-via-chatgpt-openclaw/ An OpenClaw assistant named Gila collects office grocery requests from Slack, then uses ChatGPT's Instacart app to build the cart and post the link back. Result: Gila ordered fresh office groceries through the ChatGPT Instacart app, using the card already on file, with a human approving first. - Agents: OpenClaw - Evidence: grade A (Shown, artifact linked), Blog post with a photo of the delivered groceries; no numbers - Outcome: Completed - Job type: Ops and IT - Roles: Operations - Setup: OpenClaw runs locally with a skill.md, takes requests in Slack, and types the list as plain text into ChatGPT's Instacart app, which holds the payment credentials. - Caveat: Keeps a human approval step before payment. Works only with ChatGPT apps that support in-chat checkout, and may break if OpenAI changes the app. - Prompt (written by ${BRAND}, not the author's exact prompt): Collect grocery requests from the team in [Slack channel], compile them into one shopping list, get approval from [authorized person], then use the ChatGPT Instacart app to build the cart and post the cart link and summary back. - Evidence link: https://substackcdn.com/image/fetch/$s_!Ttfh!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1c4bd726-65b6-40bc-9e5c-b67009c40618_4080x3072.jpeg - Source: https://nekuda.substack.com/p/we-connected-openclaw-with-instacart, 2026-04-09 - Published: 2026-10-13 ### Daily trending posts for content URL: https://agenticfeats.com/c/daily-trending-posts-for-content-openclaw/ OpenClaw uses the author's socialcrawl API to send a morning digest of trending posts from X, Threads and LinkedIn, used as references for writing. Result: In one week the author's Threads posts reached 100K views and gained 200 followers after using the daily digest as a writing reference. - Agents: OpenClaw - Evidence: grade B (Numbers, a specific number stated), Author-reported view and follower counts; no screenshots or repo - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Setup: OpenClaw paired with socialcrawl, a self-built API that pulls data from 21+ social platforms under one API key, scheduled to deliver trending posts every morning. - Caveat: Author built and promotes socialcrawl, and the post offers free credits. Growth is self-reported over one week, with no breakdown of other factors. - Prompt (written by ${BRAND}, not the author's exact prompt): Every morning, pull the top trending posts from the last 24 hours on [platforms] for [topic]. Send me a digest with the post text, engagement numbers and link for each. - Source: https://news.ycombinator.com/item?id=47838224 by vaaselene, 2026-04-20 - Published: 2026-10-13 ### Marketing agents on OpenClaw URL: https://agenticfeats.com/c/marketing-agents-on-openclaw/ The author built three OpenClaw agents for marketing research, long-form content structure, and copywriting drafts, and relied on them in daily work. Result: The research agent saved the author two to three hours a week; the copywriting agent removed the blank-page first draft. - Agents: OpenClaw - Evidence: grade B (Numbers, a specific number stated), Author states time saved; no repo or demo - Outcome: Completed - Job type: Marketing and Content - Roles: Marketing - Setup: Three separate agents on OpenClaw, the open-source agent platform that underpins Microsoft Scout: research, content structuring, and copywriting drafts. - Caveat: None of the agents were perfect and all needed oversight. The rest of the post covers Microsoft Scout, not the author's results. - Source: https://cgtech.com.au/microsoft-scout-always-on-agents-have-arrived-and-they-run-on-your-desktop, 2026-06-05 - Published: 2026-10-13 ### OpenAI build-fix agent on OpenClaw URL: https://agenticfeats.com/c/openai-build-fix-agent-on-openclaw/ An OpenAI engineer reports an internal OpenClaw agent, Androidclaw, working across Git, GitHub and logs to trace broken builds back to their causes. Result: Androidclaw traces broken builds to the relevant pull request and incident, and sometimes prepares and merges fixes itself. - Agents: OpenClaw - Evidence: grade C (Claim, claim only), Named OpenAI engineer quote in news article; no numbers or demo - Outcome: Completed - Job type: Coding - Roles: Developer - Industry: Software companies - Setup: Persistent OpenClaw agent with access to OpenAI codebases, plugins, company context, Git, GitHub and logging systems, governed through the OpenClaw Enterprise control plane. - Caveat: Evidence is a single staff member's account; no counts or time savings given. Agents with this access need strict permissions. - Source: https://venturebeat.com/orchestration/openclaw-launches-free-enterprise-control-plane-for-persistent-ai-agents-backed-by-openai-red-hat-and-nvidia, 2026-09-30 - Published: 2026-10-13 ### Movie ticket booking skill URL: https://agenticfeats.com/c/movie-ticket-booking-skill-openclaw/ The author had their Muse agent build a reusable skill that books movie tickets, handling calendar checks, seat choice, purchase, and a reminder to leave. Result: The agent built a skill that checks the calendar, buys the ticket via its wallet, adds the event, and sends a leave-time message. - Agents: OpenClaw, Other - Evidence: grade C (Claim, claim only), Author describes the skill steps; a screenshot is attached, no repo or booking confirmation. - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Muse agent, described as OpenClaw under the hood, using its skills feature, the built-in Link integration for payment, and calendar access. - Caveat: The source shows no completed booking, so it is unclear how well the skill works across repeated real bookings. - Prompt (written by ${BRAND}, not the author's exact prompt): Create a skill for booking movie tickets. Steps: review my calendar, confirm the date and time, confirm my seat (remember my usual preference), buy the ticket with the Link integration, add it to my calendar, and message me when it is time to leave, accounting for traffic. - Source: https://x.com/jjenglert/status/2108260489197130058 by @JJEnglert, 2026-10-08 - Published: 2026-10-13 ### Household automation add-ons for OpenClaw URL: https://agenticfeats.com/c/household-automation-add-ons-for-openclaw/ The author built OpenClaw add-ons for outbox messaging, trips, flights, and Instacart carts that run their family's household on a Mac Mini. Result: In daily production use: Instacart carts auto-staged twice weekly at 100% item accuracy, plus flight alerts and scheduled iMessages without an LLM. - Agents: OpenClaw - Evidence: grade A (Shown, artifact linked), Repo with README, screenshots of daily use, and stated accuracy figure - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: OpenClaw on a Mac Mini M4 with SQLite, Python daemons under launchd, BlueBubbles iMessage, FlightAware AeroAPI, and instacart-pp-cli. Outbox setup takes about 15 minutes. - Caveat: The agent only stages the Instacart cart; a person must tap Place Order. Earlier LLM-driven versions failed, e.g. 11 of 14 cart items wrong. - Evidence link: https://github.com/kmanan/spratt-skills - Source: https://github.com/kmanan/spratt-skills - Published: 2026-10-13 ### Overnight builds and inbox triage URL: https://agenticfeats.com/c/overnight-builds-and-inbox-triage-openclaw/ OpenClaw on a Raspberry Pi 5 triages email, builds code from a TODO list overnight, and sends daily research reports to a security consultant. Result: A VS Code extension was built overnight with 91 of 91 tests passing and a PR opened, costing about $0.35. - Agents: OpenClaw - Evidence: grade B (Numbers, a specific number stated), Author's blog post with one example build summary; no repo linked - Outcome: Completed - Job type: Business - Roles: Developer, Founder or owner - Industry: Consulting - Setup: OpenClaw on a Raspberry Pi 5 with Claude Sonnet 4, Telegram and Webchat channels, IMAP/SMTP email, cron jobs, sandboxed execution with exec approvals. - Caveat: Needed fixes for message duplication, banned Gmail account, context overflow, Docker permissions, and Telegram rate limits. Production changes require manual approval. - Source: https://simonroses.com/2026/02/my-experience-using-openclaw-a-security-professionals-journey - Published: 2026-10-13 ## Other ### Blood test PDFs to dashboard URL: https://agenticfeats.com/c/blood-test-pdfs-to-dashboard/ Claude took several unstructured blood test PDFs exported from MyChart and built an interactive browser tool to explore the results over time. Result: The tool flagged abnormal values and gave general health insights, built in about five minutes from a single prompt. - Agents: Other - Evidence: grade A (Shown, artifact linked), Live artifact link, screenshots and a video walk-through; author states five minutes - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Claude chat with the PDFs attached, plus the author's age and gender. Claude built the app as a shareable artifact. - Caveat: The author says it does not replace medical professionals or enterprise tools; the output could improve with a more detailed prompt. - Prompt (written by ${BRAND}, not the author's exact prompt): Build me an interactive browser tool for exploring my medical test result history from the attached PDFs. Highlight abnormal values and recommended actions. Make the exploration style creative and distinctive. Factor in public data relevant to a [age]-year-old [gender]. Ask any questions first. - Evidence link: https://claude.site/artifacts/2f3789fb-655a-45fb-8e67-f53d983b25b3 - Source: https://mattstockton.com/2024/07/18/unstructured-medical-data-with-llms.html, 2024-07-18 - Published: 2026-10-13 ### Flight search comparison, Atlas vs Comet URL: https://agenticfeats.com/c/flight-search-comparison-atlas-vs-comet-perplexity-comet/ The author gave ChatGPT Atlas and Perplexity Comet the same nonstop SFO to NYC flight search for a wedding trip, with airline and airport preferences. Result: Atlas returned 11 round-trip combinations with times, airports and prices. Comet gave general statements and pointed to aggregator sites instead of pulling live data. - Agents: Other, Perplexity Comet - Evidence: grade B (Numbers, a specific number stated), Author's first-hand account of one comparison; no screenshots of the results - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: Same prompt given to both agents: nonstop SFO to NYC, April 24 to 26, 2026, United, JetBlue or Alaska, JFK or LGA preferred. Atlas ran in logged-out browser mode. - Caveat: Atlas first failed with a no-data error before switching to browser mode. Eleven options was slightly too many. - Prompt (written by ${BRAND}, not the author's exact prompt): Find nonstop flights from [origin airport] to [destination city, all airports], leaving [date] after [time], returning [date] in the evening. Preferred airlines: [airlines]. Preferred arrival airports: [airports]. Show all available combinations with times and prices. - Source: https://medium.com/@amurray101/chatgpt-atlas-vs-perplexity-comet-on-flight-booking-f580f8bd6e14, 2025-10-22 - Published: 2026-10-13 ### Scout automates CRM pipeline hygiene URL: https://agenticfeats.com/c/scout-automates-crm-pipeline-hygiene/ Microsoft Scout, an agent built on OpenClaw, opened a browser, inspected CRM opportunity records, exported a summary to SharePoint, and posted compliance findings to Teams. Result: A CRM pipeline check that needed a manual export plus a separate analysis tool now runs end to end, with out-of-compliance entries flagged in Teams. - Agents: Other - Evidence: grade A (Shown, artifact linked), Newsletter write-up with screenshots; no time or volume numbers given - Outcome: Completed - Job type: Sales - Roles: Operations, Sales - Setup: Microsoft Scout desktop app (v0.22, frontier access) on OpenClaw and WorkIQ, with Teams, Outlook, SharePoint, browser control and JavaScript inspection. Runs as a scheduled automation. - Caveat: Desktop app needs the machine awake; automations pause when the device sleeps. Significant actions ask for confirmation by default. - Evidence link: https://substackcdn.com/image/fetch/$s_!vOSn!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F42208831-f65e-4805-87df-3eb369380583_1178x796.png - Source: https://copilotatwork.substack.com/p/microsoft-scout-is-not-just-another, 2026-06-06 - Published: 2026-10-13 ### 36-second video from one image URL: https://agenticfeats.com/c/36-second-video-from-one-image/ The author used Grok Imagine's Agent Mode with a single image and a simple prompt to produce a short video with matching scenes and subtitles. Result: A 36-second video with multiple camera angles, a visual storyline, and subtitles, made in about five minutes. - Agents: Other - Evidence: grade A (Shown, artifact linked), Author states five-minute build time; video attached to post - Outcome: Completed - Job type: Creative - Roles: Design - Setup: Agent Mode in Grok Imagine, starting from one image and a short prompt. - Caveat: Output is shown only as a video; the prompt and the source image are not shared. - Source: https://x.com/xfreeze/status/2078833307207483764 by @XFreeze, 2026-07-19 - Published: 2026-10-13 ### BaZi fortune agent on marketplace URL: https://agenticfeats.com/c/bazi-fortune-agent-on-marketplace/ The author built a BaZi fortune-telling agent with AI agents doing the coding, then listed it on the OKX AI Marketplace with pay-per-call USDT pricing. Result: The agent went from spec to live listing in under a week and has settled 20+ real on-chain transactions across four paid services. - Agents: Other - Evidence: grade A (Shown, artifact linked), Demo video and marketplace listing; author states 20+ transactions - Outcome: Completed - Job type: Business - Roles: Developer, Founder or owner - Setup: Deterministic calendar engine computes the chart; AI interprets it. Multilingual output in Chinese, English, Vietnamese. Payments use x402 on X Layer in USDT. - Caveat: Readings are labeled for reference and entertainment only. The transaction count is the author's own claim. - Source: https://x.com/ducktumay/status/2076933094628876322 by @Ducktumay, 2026-07-14 - Published: 2026-10-13 ### School homework to calendar URL: https://agenticfeats.com/c/school-homework-to-calendar/ Meta's Muse agent logs into a school information system through a credential vault, collects homework and due dates, and adds them to a child's calendar. Result: Set up to run daily, it creates calendar entries with a day-before reminder and a link. The author calls the first run not bad. - Agents: Other - Evidence: grade C (Claim, claim only), Author's own account; no links, repo, or screenshots - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: Muse agent with its credential vault for the school login. The author modeled the workflow formally and scheduled it to run daily. - Caveat: Only run once so far, with results still to be seen. The author shares only information they would be fine disseminating with Meta. - Prompt (written by ${BRAND}, not the author's exact prompt): Log into [school information system] using the credential vault. Read the homework posts and linked sites, list each assignment with its due date, add each to [child]'s calendar, and set a reminder the day before with a link to the entry. Run daily. - Source: https://news.ycombinator.com/item?id=49796524 by fnordpiglet, 2026-09-22 - Published: 2026-10-13 ### ChatGPT Dots day-one test URL: https://agenticfeats.com/c/chatgpt-dots-day-one-test/ Julian Goldie tested OpenAI's new ChatGPT Dots agent on launch day with research, a website build, computer use, and a restaurant booking. Result: Research, site build and restaurant booking worked, but computer use was very slow and the call button failed. - Agents: Other - Evidence: grade A (Shown, artifact linked), Author's video demo plus a table of per-task test results - Outcome: Partial - Job type: Research - Roles: Marketing - Setup: ChatGPT Pro account with a Dot using its own cloud computer; apps connected through the plugins section, with Zapier MCP suggested by the Dot. - Caveat: Day-one rollout bugs; access limited by plan and region, and the author would not switch from Claude, Grok Bot or Hermes yet. - Evidence link: https://www.youtube.com/watch?v=RJEzrQHv0_o - Source: https://juliangoldie.com/chatgpt-dots, 2026-09-30 - Published: 2026-10-13 ### AI-built online store URL: https://agenticfeats.com/c/ai-built-online-store/ Ryan Robinson used Bluehost's AI store builder to generate a retail store for Little River Studios from one product description. Result: The author launched a store taking real payments in one sitting and ran a test order. Five agents now watch orders and stock. - Agents: Other - Evidence: grade A (Shown, artifact linked), Blog post with video walkthrough and screenshots; sponsored by Bluehost - Outcome: Completed - Job type: Business - Roles: Founder or owner - Industry: Retail shops - Setup: Bluehost AI store builder on WordPress and WooCommerce. One prompt described the products; checkout was added by chat request; Stripe, taxes, and shipping were set in the dashboard. - Caveat: Starter products are placeholders, checkout must be requested, and edits run on a monthly credit system. The post is sponsored by Bluehost. - Prompt (written by ${BRAND}, not the author's exact prompt): Build an online store for [business name], a curated shop selling [product types, e.g. desk gear and studio equipment]. Include category pages, filtered product pages, and product details. - Source: https://ryrob.com/run-online-store-with-ai, 2026-09-30 - Published: 2026-10-13 ### Backlink outreach for Promptslove URL: https://agenticfeats.com/c/backlink-outreach-for-promptslove/ The author's ChatGPT Dot, Jamie, researched publishers, drafted pitches, sent outreach and form submissions, and tracked results for Promptslove's digital PR campaign. Result: Jamie reported 13 outreach actions and zero live links early on; the author later verified the first live nofollow directory backlink. - Agents: Other - Evidence: grade A (Shown, artifact linked), Public Baserow grid of the live listing plus conversation screenshots; actions mostly agent-reported - Outcome: Partial - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Saas - Setup: OpenAI Dots in ChatGPT with a connected business mailbox, contact forms, and a public Baserow tracking grid. The author specified the approved business sender and authorized outreach. - Caveat: The first link is a nofollow directory listing with editorial approval unconfirmed. Many sends are agent-reported only, and one send had no confirmation. - Prompt (written by ${BRAND}, not the author's exact prompt): Act as my digital PR agent for [site]. Find [number] qualified prospects, each with the exact page, audience fit, resource to pitch, and an official contact route. Send only from [approved business sender]. Use only claims from the approved claims ledger. Ask me before anything outside these limits. - Evidence link: https://baserow.io/public/grid/OH9K9Io1VgQBMkTe0bnqTgnDvoqOA9o-HDeVymrX298 - Source: https://promptslove.com/blog/openai-dots-guide, 2026-09-30 - Published: 2026-10-13 ### Email calendar scheduling assistant URL: https://agenticfeats.com/c/email-calendar-scheduling-assistant/ The author's AI assistant reads forwarded emails in a labeled Gmail inbox, emails recipients to coordinate times, and adds Zoom bridges. Result: After a few months of iterations, the author has a working assistant, open sourced so others can have their AI configure it for their needs. - Agents: Other - Evidence: grade A (Shown, artifact linked), Author says it works; open source project page linked - Outcome: Partial - Job type: Finance and Admin - Roles: Operations - Setup: Forward or create meeting-request emails into a Gmail labeled inbox. Runs on OpenAI's Dot; Zoom is used for the bridge. Open source project to download and configure. - Caveat: Group scheduling is untested. Running it through Codex was problematic. The author's version has extra complexity others may not need. - Evidence link: https://projects.eddiesoehnel.com/adminprojects/calendar-scheduling-OPEN - Source: https://x.com/edsoehnel/status/2107574500984287570 by @edsoehnel, 2026-10-06 - Published: 2026-10-13 ### Microsoft 365 on Linux URL: https://agenticfeats.com/c/microsoft-365-on-linux/ A Nix flake runs Microsoft 365 and Office Home 2024 on Linux through umu-launcher and GE-Proton, with shims fixing Wine crashes. Result: Word starts and draws its ribbon, OneNote signs in and saves edits, and a retail Home 2024 key activated through account sign-in. - Agents: Other - Evidence: grade A (Shown, artifact linked), Repo README with status notes and a fixes table; no screenshots - Outcome: Partial - Job type: Ops and IT - Roles: Developer - Setup: Nix flake with umu-launcher and GE-Proton or ProtoSoda, winetricks fonts and DLLs, plus custom shims for licensing, ole32 and d2d1. - Caveat: Office LTSC and volume-license keys do not activate under Wine. Excel, PowerPoint and Outlook were untested at last status. - Evidence link: https://github.com/Tombert/office365_flake - Source: https://github.com/Tombert/office365_flake - Published: 2026-10-13 ## Cursor ### NSF project summary drafting URL: https://agenticfeats.com/c/nsf-project-summary-drafting-cursor/ The author used Cursor in Agent mode with Claude 3.7 Sonnet and two rules files to draft an NSF Project Summary from two papers. Result: The draft was usable but weaker than accepted NSF summaries; thinking mode helped, though it began hallucinating and misreading the sources. - Agents: Cursor - Evidence: grade C (Claim, claim only), Author's forum post describing the experiment; screenshot of the folder setup only - Outcome: Partial - Job type: Research - Roles: Education - Industry: Universities - Setup: Cursor Agent mode, claude-3.7-sonnet, one rules file for NSF requirements, one for academic writing, section markdown files, and a reference folder with two papers as text. - Caveat: The writing rule was not reliably picked up unless @-mentioned. Thinking mode was needed, and the output still needed heavy critique. - Source: https://forum.cursor.com/t/anyone-use-cursor-for-academic-research-writing/57393, 2025-02-28 - Published: 2026-10-13 ### Blog drafts from Obsidian notes URL: https://agenticfeats.com/c/blog-drafts-from-obsidian-notes-cursor/ The author set Cursor to imitate their note.com writing style and read their Obsidian vault, then had it draft blog articles quoting past reading highlights. Result: Cursor drafted an article on their reading trends in their style, citing Kindle highlights from the vault. The author edited it by hand. - Agents: Cursor - Evidence: grade C (Claim, claim only), Author's own account with an excerpt of the generated article; no repo. - Outcome: Partial - Job type: Marketing and Content - Roles: Developer, Personal - Setup: Cursor with custom User Rules describing the author's writing style, plus a local Obsidian vault of Markdown files (Kindle highlights, event notes, daily notes) referenced by directory path. - Caveat: The generated text still feels slightly off and needed substantial manual editing. The author says it can't be left entirely to the AI. - Prompt (written by ${BRAND}, not the author's exact prompt): Follow my writing style as described in [style notes or blog URL]. Use the Markdown notes in [vault path] as your knowledge base. When quoting a book, use a blockquote with the book title and author. Draft an article on [topic]. - Source: https://note.com/dora_e_m/n/nc6abb651ea77?hl=en, 2025-03-01 - Published: 2026-10-13 ### Terraform via MCP in Cursor URL: https://agenticfeats.com/c/terraform-via-mcp-in-cursor/ The author used Cursor with the tfmcp MCP server to run terraform init, plan and apply, provisioning a Postgres database, schema and app role locally. Result: Provisioning worked after two fixes. The MCP hid the real errors, so apply ran through Cursor's Run tool, which suggested the fix. - Agents: Cursor - Evidence: grade A (Shown, artifact linked), Blog post with config, Terraform code and debugging screenshots; no stated time or money saved - Outcome: Partial - Job type: Ops and IT - Roles: Developer, Operations - Industry: Consulting - Setup: Cursor agent mode with the tfmcp MCP server, configured in mcp.json with the Terraform directory and PATH. Terraform code used the cyrilgdn/postgresql provider. - Caveat: The MCP did not surface Terraform errors or confirm apply success. The author calls MCP workflows very early and flags security risks of adding MCPs. - Source: https://masterpoint.io/blog/using-mcps-to-run-terraform, 2025-04-17 - Published: 2026-10-13 ### Pricing page via background agent URL: https://agenticfeats.com/c/pricing-page-via-background-agent-cursor/ Cursor's background agent built a pricing landing page from a short spec, working on its own branch and opening a pull request. Result: The agent finished the page in about 5 minutes; the author reviewed the PR, tested locally, and merged it to production. - Agents: Cursor - Evidence: grade B (Numbers, a specific number stated), Author states 5 minutes and merge; screenshots of the PR, no repo - Outcome: Completed - Job type: Coding - Roles: Developer, Founder or owner - Setup: Cursor background agent mode, given a small ticket-style spec. It works on a new branch and creates a pull request. - Caveat: The author has only tried small, easy features so far. BugBot reviews are pricey because they use Max mode. - Prompt (written by ${BRAND}, not the author's exact prompt): Write a short spec for [feature] like a ticket for a junior developer: goal, page sections, design notes, and constraints. Build it on a new branch and open a pull request. - Source: https://thepromptwarrior.com/p/how-these-cursor-updates-will-save-you-hours, 2025-06-07 - Published: 2026-10-13 ### Newsletter and posts in own voice URL: https://agenticfeats.com/c/newsletter-and-posts-in-own-voice-cursor/ The author built a Cursor workflow turning raw notes into a newsletter, LinkedIn posts and tweet threads in their own voice. Result: A messy one-line note became a 2000-3000 word newsletter plus LinkedIn and Twitter drafts; the post itself says it was generated this way and published. - Agents: Cursor - Evidence: grade A (Shown, artifact linked), Author states the newsletter was generated by the system; demo video linked - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Setup: Cursor with Python scripts, Selenium to scrape LinkedIn posts of marketing leaders, feed parsing of newsletters, and style learning from 32+ of the author's own posts. Output saved to Git folders. - Caveat: Author is the only source of the quality claim; no engagement numbers or time savings are given, and LinkedIn scraping needs browser automation. - Evidence link: https://www.youtube.com/watch?v=sJEqbOS_s28 - Source: https://newsletter.qback.ai/p/ai-marketing-content-generator-cursor, 2025-08-25 - Published: 2026-10-13 ### Figma design to HTML URL: https://agenticfeats.com/c/figma-design-to-html-cursor/ Cursor, connected to Figma's Dev Mode MCP, turned a selected Figma design into HTML and CSS from one prompt containing the design URL. Result: One prompt produced the thumbnail section with accurate colours, layout and button hover states, and it beat the earlier Lovable attempt built from static images. - Agents: Cursor - Evidence: grade A (Shown, artifact linked), Author describes the output with screenshots; no repo or live site linked - Outcome: Partial - Job type: Coding - Roles: Design, Developer - Setup: Figma Dev Mode MCP linked to Cursor on its free Pro trial. A local Git repo synced to GitHub was opened in Cursor, then the Figma selection URL was pasted into the prompt. - Caveat: It missed the font and font weights, likely because Rotis is not a free web font. More tweaking is still needed. - Prompt (written by ${BRAND}, not the author's exact prompt): Using the Figma design at [Figma Dev Mode URL], build this component in HTML and CSS in this repository. - Source: https://designphilosophy.substack.com/p/getting-remarkable-results-quickly, 2025-08-05 - Published: 2026-10-13 ### Cursor workflows for DevOps URL: https://agenticfeats.com/c/cursor-workflows-for-devops/ Cursor was used by a DevOps engineer with rules files, MCP plugins, and auto-run commands to refactor configs, debug builds, and migrate infrastructure code. Result: The author migrated 10+ Lambda functions from Node.js 14 to 20 in one pass and fixed a broken docker-compose file by auto-retrying until clean. - Agents: Cursor - Evidence: grade C (Claim, claim only), Author's first-person account with example prompts; no repo or demo. - Outcome: Completed - Job type: Ops and IT - Roles: Developer, Operations - Setup: Cursor with a .cursorrules file, global rules, Kubernetes and GitHub MCP plugins, a .cursorignore file, notepads for CI/CD templates, and agent auto-run commands enabled. - Caveat: Cursor once hallucinated secrets from a .env file until it was ignored. AI changes can go rogue, so the author commits often and saves checkpoints. - Prompt (written by ${BRAND}, not the author's exact prompt): Fix this Dockerfile for a [language] app. Base image: [image]. Run as non-root user. Add a multistage build for production. Keep size under [size]. Validate with 'docker build --no-cache'. - Source: https://medium.com/@soodrajesh/advanced-tips-and-tricks-for-cursor-ide-7dc236ff9870, 2025-08-23 - Published: 2026-10-13 ### Consulting jargon minigame collection URL: https://agenticfeats.com/c/consulting-jargon-minigame-collection-cursor/ Alvin Leung used ChatGPT to refine a game concept, then Cursor Agent mode to generate a four-minigame Python tkinter game themed on consulting life. Result: A playable game with four minigames, persistent high scores, and no external dependencies was built in a couple of hours, with source code on GitHub. - Agents: Cursor - Evidence: grade A (Shown, artifact linked), GitHub repo and demo GIF linked; author states a couple of hours - Outcome: Completed - Job type: Creative - Roles: Personal - Setup: ChatGPT to refine the concept and write a system prompt, then Cursor Agent mode to build a scene-based Python 3.13 tkinter app with JSON high scores. - Caveat: The author says AI handled implementation, not design. Balance tweaks were done by hand. Feedback came from only three colleagues. - Prompt (written by ${BRAND}, not the author's exact prompt): Create a [theme]-themed game with [number] minigames: [list of minigame ideas]. Use Python with tkinter, keep it simple, and focus on time-based scoring with penalties for mistakes. - Evidence link: https://github.com/NotAlvin/consulting-chaos - Source: https://medium.com/@leung.alvin/how-i-built-a-complete-game-in-one-day-using-chatgpt-and-agent-mode-in-cursor-1c39b336134b, 2025-09-29 - Published: 2026-10-13 ### App review analysis for PMs URL: https://agenticfeats.com/c/app-review-analysis-for-pms-cursor/ Cursor analyzed a CSV of a couple thousand Walmart App Store reviews and returned takeaways with ASCII diagrams. Result: In 30 seconds it returned top complaints ranked by frequency, a severity versus frequency matrix, and a user journey map showing where frustration peaks. - Agents: Cursor - Evidence: grade A (Shown, artifact linked), Guest post with screenshots and a recorded lightning lesson; no repo - Outcome: Completed - Job type: Research - Roles: Product - Setup: Cursor with the review CSV in the project. For brainstorming, an AGENTS.md file the agent generated from the cloned codebase. Models mentioned: Composer 1, Claude Opus 4.5, Gemini 3 Flash. - Caveat: Review data was public and the demo was shown for teaching purposes, not tied to a shipped product change. - Prompt (written by ${BRAND}, not the author's exact prompt): Help me analyze the [review data file] and give me high level takeaways. Break the data into clear, concise insights in as few words as possible. Give me high level diagrams using ASCII. - Evidence link: https://maven.com/p/caee21/cursor-for-product-managers - Source: https://amankhan1.substack.com/p/cursor-for-product-managers, 2026-01-05 - Published: 2026-10-13 ### Fixing a slow page with indexes URL: https://agenticfeats.com/c/fixing-a-slow-page-with-indexes-cursor/ Cursor's agent listed likely causes of a production-only slowdown, traced NestJS database queries, and recommended which database indexes to add. Result: After the author added the suggested indexes, page load performance improved significantly in production. - Agents: Cursor - Evidence: grade C (Claim, claim only), Author's own account; no repo, numbers, or demo - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Cursor agent working inside the existing NestJS codebase, with the agent reading the database queries the app makes. - Caveat: The author warns that agent output needs careful review. Their python unit tests failed on bad mocks and needed manual fixes. - Prompt (written by ${BRAND}, not the author's exact prompt): Our [system] is slow in production but fast in dev. List possible causes and steps to fix them, based on the database queries this codebase makes and which indexes they should use. - Source: https://trevorarnold.substack.com/p/december-2025-lessons, 2026-01-05 - Published: 2026-10-13 ### Hundreds of agents build a browser URL: https://agenticfeats.com/c/hundreds-of-agents-build-a-browser-cursor/ Cursor ran hundreds of concurrent coding agents, split into planners and workers, on one project for close to a week to build a web browser. Result: The agents wrote over 1 million lines of code across 1,000 files, and a three-week Solid to React migration reached +266K/-193K edits. - Agents: Cursor - Evidence: grade A (Shown, artifact linked), Public GitHub repos for the browser and other experiments, plus a PR screenshot - Outcome: Partial - Job type: Coding - Roles: Developer - Setup: Planner agents create tasks and spawn sub-planners, workers complete tasks and push to one branch, and a judge agent decides whether to continue. GPT-5.2 models were used. - Caveat: The Solid to React migration still needs careful review. Agents sometimes run too long and need periodic fresh starts to combat drift. - Evidence link: https://github.com/wilsonzlin/fastrender - Source: https://cursor.com/blog/scaling-agents, 2026-01-14 - Published: 2026-10-13 ### Reverse engineering a Go binary URL: https://agenticfeats.com/c/reverse-engineering-a-go-binary-cursor/ Cursor, connected to Ghidra through a GhidraMCP server, analyzed a stripped Golang HTTP binary and identified its libraries, API routes, and file structure. Result: The agent rebuilt readable Go source from the stripped binary, but got some route paths wrong, such as /my-messages instead of /me/messages. - Agents: Cursor - Evidence: grade A (Shown, artifact linked), Blog post with screenshots and recovered code; author states errors - Outcome: Partial - Job type: Coding - Roles: Developer - Setup: Cursor with the GhidraMCP bridge. Ghidra with the Golang Analyzer Extension to restore symbols in the stripped binary. Any MCP-capable agent would also work. - Caveat: Tested on a small demo server, not a large program. Some routes were wrong, and parts looked skipped lazily. - Prompt (written by ${BRAND}, not the author's exact prompt): I am reverse engineering a Golang binary. Use Ghidra MCP to tell me what kind of program it is and what functions it has. Then convert the decompiled C back into Go. - Source: https://blog.huli.tw/2026/03/01/en/reverse-engineering-with-ai-ghidra-mcp, 2026-03-01 - Published: 2026-10-13 ### Dependency and framework upgrade URL: https://agenticfeats.com/c/dependency-and-framework-upgrade-cursor/ Cursor listed outdated packages by upgrade difficulty, then upgraded them in stages, including Remix v2 to React Router v7 on a 42k-line site. Result: The author upgraded kentcdodds.com across all dependency tiers, finishing a Remix v2 to React Router v7 migration with routing matching the original structure. - Agents: Cursor - Evidence: grade A (Shown, artifact linked), Author's blog post with prompts and a linked PR; no hard time saved - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Cursor (Composer 1.5) plus a long-running background agent for the big migration. The agent searched migration docs. The project's existing tests and docs guided the work. - Caveat: The agent struggled with the routing convention change and needed a corrective prompt. Good existing tests and docs made it work. - Prompt (written by ${BRAND}, not the author's exact prompt): Upgrade us from [framework A] to [framework B]. Before doing this, look up relevant migration docs and upgrade guides. Check that our framework-specific dependencies work with the new version, and find alternatives if they do not. Replace patch files with [simpler approach]. - Evidence link: https://github.com/kentcdodds/kentcdodds.com/pull/658 - Source: https://kentcdodds.com/blog/how-i-used-cursor-to-migrate-frameworks - Published: 2026-10-13 ### Cursor as Obsidian vault assistant URL: https://agenticfeats.com/c/cursor-as-obsidian-vault-assistant/ Cursor reads the author's whole Obsidian vault, reviews note structure, writes daily done notes from work outside Obsidian, and answers questions about their notes. Result: The author says the setup saves hours every week and keeps daily notes consistent, though no measured figures are given. - Agents: Cursor - Evidence: grade C (Claim, claim only), Author's own account with workflow diagrams; no repo or measured data - Outcome: Completed - Job type: Setup and Meta - Roles: Personal - Setup: Cursor is given access to the full Obsidian vault. The author asks it to check notes, suggest links, log work, and query note properties. - Caveat: The author reviews every suggestion before applying it. Time saved is self-reported with no numbers, and Cursor is not strictly required. - Prompt (written by ${BRAND}, not the author's exact prompt): Review the structure of my notes in [vault folder]. Suggest missing links, better tags, and inconsistent frontmatter, but do not create new notes. - Source: https://beatriz.page/2026-01-23-how-i-use-obsidian-and-cursor-together - Published: 2026-10-13 ### Cursor for data science scorecard URL: https://agenticfeats.com/c/cursor-for-data-science-scorecard/ The author used Cursor on a real data science project in Python and graded it across general coding and data tasks. Result: Prototyping, code completion, commenting and documentation earned A grades; web scraping, code tidying and getting started got D; dataframe manipulation and code generation got C. - Agents: Cursor - Evidence: grade C (Claim, claim only), Author's first-hand scorecard from a real project; no repo linked - Outcome: Partial - Job type: Coding - Roles: Developer - Setup: Cursor with Claude as the engine, on a paid subscription, working on Python, Pandas and Streamlit code with GitHub integration. - Caveat: Technical and dataframe code needed manual review or hand rewriting, Cursor deleted some data files via GitHub, and there is no notebook support. - Source: https://blog.engora.com/2025/06/cursor-for-data-science-scorecard.html - Published: 2026-10-13 ## Devin ### Unanswered email tracker with Devin URL: https://agenticfeats.com/c/unanswered-email-tracker-with-devin/ Devin wrote a Gmail function to find emails needing a reply, built a front end, tested it locally, and deployed it to a public link. Result: Within one night Devin produced a working unanswered-emails interface using the Gmail API, after the author wrote the authentication code themselves. - Agents: Devin - Evidence: grade A (Shown, artifact linked), Screenshots of Devin's work and the deployed interface; no repo - Outcome: Partial - Job type: Coding - Roles: Developer, Founder or owner - Setup: Devin with the Gmail API. Author handled authentication in Cursor. Devin then moved on to a Supabase backend so emails can be ignored and remembered. - Caveat: Devin struggled with authentication, so the author wrote it themselves. It also worked at human speed and was slower than the author on Supabase setup. - Prompt (written by ${BRAND}, not the author's exact prompt): Create a function that finds emails from the past [time window] that I still need to respond to, and make sure they are in my Primary inbox. - Source: https://jamesin.substack.com/p/hello-world-meet-devin, 2024-12-11 - Published: 2026-10-13 ### One month testing Devin URL: https://agenticfeats.com/c/one-month-testing-devin/ Devin was connected to the RunLLM GitHub repo and given small bug fixes and one larger full-stack task over a month. Result: Small scoped fixes took about five minutes to test and merge versus roughly an hour by hand; the full-stack task failed and was rewritten. - Agents: Devin - Evidence: grade B (Numbers, a specific number stated), Author's firsthand account with time estimates; no repo or demo - Outcome: Partial - Job type: Coding - Roles: Developer, Founder or owner - Industry: Software companies - Setup: Devin connected to the GitHub repo, scanned it to infer build steps, and opened pull requests. Tasks were typed into its chat interface. - Caveat: Devin struggled with a multi-layer task, ignored taught practices, and its code was unworkable. Backend work went worse than frontend. - Source: https://frontierai.substack.com/p/one-month-of-using-devin, 2025-01-23 - Published: 2026-10-13 ### Third-party token API endpoint URL: https://agenticfeats.com/c/third-party-token-api-endpoint-devin/ Devin AI planned and built a backend API endpoint for generating third-party user access tokens, then opened a GitHub pull request. Result: After 2-3 review rounds the PR was production-ready and merged. The task took about 1 hour versus roughly 15 minutes coding it alone. - Agents: Devin - Evidence: grade A (Shown, artifact linked), Author states timings and shows a merged PR screenshot; no repo link - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Devin was given the third-party API docs, the target microservice name, and instructions to reuse the existing CDK secrets logic for the API key. Review feedback went in as GitHub PR comments. - Caveat: First pass missed the API key and style rules; the author fixed these in 2-3 review rounds. - Prompt (written by ${BRAND}, not the author's exact prompt): Implement the backend API based on this documentation [link to third-party docs]. Add it to [microservice name]. Use the existing CDK secrets logic to store the third-party API key. - Source: https://dev.to/prpatel05/a-real-world-coding-story-devin-ai-as-my-co-pilot-548, 2025-02-14 - Published: 2026-10-13 ### Docs maintenance with Devin URL: https://agenticfeats.com/c/docs-maintenance-with-devin/ Devin handled a docs team's recurring work: weekly changelog setup, GitHub issue and PR fixes, a multi-file term rename, and code explanations. Result: Devin found nearly every old term in a docs-wide rename, but the edits still needed 20+ manual commits. - Agents: Devin - Evidence: grade B (Numbers, a specific number stated), Team write-up with one stated figure; no repo or demo linked - Outcome: Partial - Job type: Marketing and Content - Roles: Developer, Product - Industry: Software companies - Setup: Devin with a playbook for the weekly changelog setup. It is pointed at GitHub issues and PRs and opens pull requests against the docs repo. - Caveat: Find-and-replace edits produced awkward phrasing like "instant instant restore", and the team made over 20 manual polish commits afterward. - Prompt (written by ${BRAND}, not the author's exact prompt): Set up a changelog for this [day]. Review issue [issue link], find the relevant parts of the docs, and open a PR with the updates. - Source: https://neon.com/blog/ai-workflows-for-docs-devin, 2025-06-23 - Published: 2026-10-13 ### City dashboard from blank repo URL: https://agenticfeats.com/c/city-dashboard-from-blank-repo-devin/ Devin took a blank GitHub repo and built a City Explorer dashboard with a FastAPI backend and React frontend, then deployed it. Result: A working, deployed app that shows images, weather, and news for a city, finished in roughly 90 minutes from an empty repo. - Agents: Devin - Evidence: grade C (Claim, claim only), Author's first-hand account; no repo or live URL linked - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Devin Core plan. Each session runs on a cloud Ubuntu 22.04 VM. Input was a GitHub repo URL; Devin used its in-browser VS Code environment. - Caveat: Integrations were placeholders. Runs only in Devin's cloud, and ACU billing counts idle and thinking time. - Prompt (written by ${BRAND}, not the author's exact prompt): Build a [app name] in this empty GitHub repo: a dashboard that shows [data sources] for a given [input]. Use [backend] and [frontend], open a pull request, and deploy to a public URL. - Source: https://medium.com/@sambhavgaur_70582/copilot-cursor-or-devin-my-hands-on-weekend-with-ai-that-codes-and-deploys-acc708e802b3, 2025-08-13 - Published: 2026-10-13 ### Devin builds Devin URL: https://agenticfeats.com/c/devin-builds-devin/ Devin handles Cognition's own engineering work across Slack, Linear, and the web app: PRs, code review, bug triage, debugging, and design system audits. Result: Cognition merged 659 Devin PRs into its own codebase in one week, up from 154 in its best week of 2025. - Agents: Devin - Evidence: grade B (Numbers, a specific number stated), Author states PR counts; screenshots of Slack and Linear flows, no repo - Outcome: Completed - Job type: Coding - Roles: Developer - Industry: Software companies - Setup: Repos added to Devin and indexed. Slack and Linear integrations, a triage playbook fired by the Bug label, Datadog MCP, read-only database access. - Caveat: Written by a Cognition employee about their own product, so there is no independent check. Devin Review and Autofix are the company's own tools. - Prompt (written by ${BRAND}, not the author's exact prompt): Triage bug [ticket]: read the report, search the codebase for relevant code paths, check recent git history, then post root cause, affected files, and a suggested fix back to [issue tracker]. - Source: https://nader.substack.com/p/how-cognition-uses-devin-to-build, 2026-02-11 - Published: 2026-10-13 ### Devin building Devin URL: https://agenticfeats.com/c/devin-building-devin/ Cognition's team uses Devin across Slack, Linear, web, CLI and API to write, review, triage and debug code in its own codebase. Result: Cognition merged 659 Devin-authored PRs into its own codebase in one week, up from 154 in its best week of 2025. - Agents: Devin - Evidence: grade A (Shown, artifact linked), Vendor blog states weekly PR counts; screenshots, no independent repo - Outcome: Completed - Job type: Coding - Roles: Developer, Product - Industry: Software companies - Setup: Repos added to Devin and indexed. Used via Slack, Linear/Jira tags, web app, with playbooks and MCP integrations such as Datadog, Sentry and Redshift. - Caveat: Written by the vendor about its own product; the post gives no detail on review effort or failure rates. - Source: https://cognition.com/blog/how-cognition-uses-devin-to-build-devin, 2026-02-27 - Published: 2026-10-13 ### Devin one-week coding review URL: https://agenticfeats.com/c/devin-one-week-coding-review/ Devin was given a week of real coding tasks, from a simple Express API to pagination, and reviewed on speed and accuracy. Result: Devin built a three-endpoint Express API in about 4 minutes, but chose the wrong pagination style on a real codebase and needed review and refactoring. - Agents: Devin - Evidence: grade C (Claim, claim only), Author's first-hand week-long review; no repo or demo shown - Outcome: Partial - Job type: Coding - Roles: Developer - Setup: Devin on the Teams plan, with its own browser, terminal and editor in a sandbox. Tasks were given in plain English. - Caveat: Slow at 20-30 minutes for a 10-minute task, got stuck in debug loops twice, and costs $500/month. - Prompt (written by ${BRAND}, not the author's exact prompt): Add pagination to the [endpoint] endpoint. Support [page and limit] query params and return the total count in the response. Follow the pagination pattern already used in this codebase. - Source: https://aimadetools.com/blog/devin-one-week-review, 2026-04-17 - Published: 2026-10-13 ### Devin cloud testing of PRs URL: https://agenticfeats.com/c/devin-cloud-testing-of-prs/ Devin writes a test plan, runs the app on a cloud VM with computer use, and returns annotated test videos and reports. Result: Test runs approved per day on Devin more than doubled in a couple of months, and engineers run 10 to 20 Devins in parallel. - Agents: Devin - Evidence: grade B (Numbers, a specific number stated), Vendor's own account with GIFs; no independent customer named - Outcome: Completed - Job type: Coding - Roles: Developer - Industry: Software companies - Setup: Devin's computer use tools on a cloud VM, a code-grounded test plan, login scripts saved as repo testing skills, and a YAML blueprint snapshot. - Caveat: Timing issues can miss transient UI like toasts, and models sometimes run JavaScript instead of clicking through the UI. First setup often needs credentials. - Source: https://cognition.ai/blog/testing-development, 2026-05-29 - Published: 2026-10-13 ### Marketing site from Figma URL: https://agenticfeats.com/c/marketing-site-from-figma-devin/ Devin AI connected to a Figma file through MCP and built a WordPress marketing site with easy content management overnight. Result: The unfinished output is work the author would have quoted at about €2.5k, built in one overnight run. - Agents: Devin - Evidence: grade B (Numbers, a specific number stated), Author states the quote-equivalent value; no repo or screenshot - Outcome: Partial - Job type: Coding - Roles: Developer - Setup: Devin AI linked to the Figma design file via MCP, targeting a WordPress front end with simple content management. - Caveat: The site is not finished yet. The €2.5k figure is the author's own quote estimate. - Source: https://x.com/bnistordev/status/2104882458671907327 by @bnistordev, 2026-09-29 - Published: 2026-10-13 ### Devin as feedback triage bot URL: https://agenticfeats.com/c/devin-as-feedback-triage-bot/ Devin triaged every app feedback and feature request in a dedicated Slack channel, with read-only access for debugging, and let non-technical staff coauthor PRs. Result: Users resolved issues by talking to Devin, engineers only merged when needed, and Devin Review caught bugs. - Agents: Devin - Evidence: grade C (Claim, claim only), Author's firsthand account; no repo, numbers, or artifact - Outcome: Completed - Job type: Ops and IT - Roles: Developer, Product - Setup: Devin connected to a Slack feedback channel, with read-only permissions so it could debug async flows and infrastructure issues in depth. - Caveat: Expensive, especially Devin Review. The author's team paid the premium anyway because time to value was fast. - Source: https://x.com/sidstronaut/status/2105417679691485630 by @sidstronaut, 2026-09-30 - Published: 2026-10-13 ### Month of Devin tasks URL: https://agenticfeats.com/c/month-of-devin-tasks/ Devin was given more than 20 real coding, research, and code-modification tasks over a month by an Answer.AI team to test its autonomy. Result: Out of 20 tasks, Devin had 14 failures, 3 successes, and 3 inconclusive results, with no pattern predicting which would work. - Agents: Devin - Evidence: grade B (Numbers, a specific number stated), Team write-up with counts and an appendix of tasks; no output repo - Outcome: Failed, with a lesson - Job type: Coding - Roles: Developer - Industry: Software companies - Setup: Devin, driven through Slack and its web interface, running in its own Docker environment with browser, editor, and shell. - Caveat: Failed tasks often took days. Devin pushed ahead on impossible tasks and produced overly complex code instead of flagging limits. - Source: https://answer.ai/posts/2025-01-08-devin.html - Published: 2026-10-13 ### Month-long Devin trial URL: https://agenticfeats.com/c/month-long-devin-trial/ Devin was tested by the Answer.AI team on 20+ real coding, research, and code-modification tasks over a month through Slack. Result: Out of 20 tasks, Devin had 14 failures, 3 successes, and 3 inconclusive results, with no predictable pattern for which would work. - Agents: Devin - Evidence: grade B (Numbers, a specific number stated), Detailed blog post with task appendix and counts; no output repo - Outcome: Partial - Job type: Coding - Roles: Developer - Setup: Devin used through Slack and its web interface, running in its own Docker environment with browser, editor, and shell. - Caveat: Tasks often took days, Devin pushed ahead on impossible tasks like a multi-app Railway deployment, and produced overly complex code. - Source: https://answer.ai/posts/2025-01-08-devin - Published: 2026-10-13 ### Vue 2 to Vue 3 migration URL: https://agenticfeats.com/c/vue-2-to-vue-3-migration-devin/ Devin was asked to migrate a production Electron app from Vue 2 to Vue 3 with Composition API, script setup, and Pinia. Result: Devin opened a PR in about four hours and got the build passing, but runtime issues remained and the author gave up. - Agents: Devin - Evidence: grade C (Claim, claim only), Author's detailed first-hand account with chat excerpts; no repo of the migration - Outcome: Failed, with a lesson - Job type: Coding - Roles: Developer - Setup: Devin with GitHub repo access only. No Slack integration, no teammates. The author told it to commit every five components and to proceed on package changes without asking. - Caveat: Devin claimed the build passed when it failed, needed days of correction, and the code still did not work at runtime. - Prompt (written by ${BRAND}, not the author's exact prompt): Upgrade this project from [old framework version] to the latest [framework] version. After migration, it should use [target API style]. After every five components updated, commit and push. Proceed with package changes without asking. - Source: https://seankearney.com/post/devinai-vuejs-migration - Published: 2026-10-13 ## Manus ### Marketing plan and website build URL: https://agenticfeats.com/c/marketing-plan-and-website-build-manus/ Manus created a marketing plan with audience personas and HubSpot workflow steps for the author's company, then built a multi-page website from that plan. Result: Manus built, tested, and published a multi-page website to a temporary server in under 10 minutes, which the author could not have built themselves. - Agents: Manus - Evidence: grade B (Numbers, a specific number stated), Author's first-hand account with screenshots; no live site or repo shared - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Marketing agencies - Setup: Manus early-access account. The author gave parameters, answered clarifying questions, uploaded context, and accepted its offer to turn the plan into a website. - Caveat: The author declined permanent publishing, unsure where Manus would host it, and flags brand, security, and legal risks. - Prompt (written by ${BRAND}, not the author's exact prompt): Create a marketing plan for my company, [company name]. Base it on my current marketing technology stack, [tools]. Include target audience personas and step-by-step implementation instructions for [CRM]. Ask me any clarifying questions first. - Source: https://loopagency.ai/insights/manus-and-the-future-of-ai-agents, 2025-04-09 - Published: 2026-10-13 ### Manus tested on research tasks URL: https://agenticfeats.com/c/manus-tested-on-research-tasks/ Manus researched China tech reporters, NYC two-bedroom listings, and Innovators Under 35 candidates for an MIT Technology Review reporter, who steered it with feedback. Result: It produced a 30-journalist list, a tiered NYC apartment list in under half an hour, and only five solid nominees out of fifty requested. - Agents: Manus - Evidence: grade B (Numbers, a specific number stated), Published hands-on review with task-by-task outcomes and timings; no output artifacts linked - Outcome: Partial - Job type: Research - Roles: Marketing - Setup: Manus general agent with invite access. Uses Claude 3.5 Sonnet and fine-tuned Qwen models. The author watched the Manus's Computer window and took over for captchas. - Caveat: Frequent crashes and server overload, paywall and captcha blocks, and a three-hour nomination task that stalled. First answers were lazy until the author pushed back. - Prompt (written by ${BRAND}, not the author's exact prompt): Compile a list of [number] notable [type of person] covering [topic]. For each, give their current outlet and notable work. Be consistent and thorough, and check for recent employer changes. - Source: https://technologyreview.com/2025/03/11/1113133/manus-ai-review, 2025-03-11 - Published: 2026-10-13 ### Synthetic AGI attitudes survey study URL: https://agenticfeats.com/c/synthetic-agi-attitudes-survey-study-manus/ Manus planned a survey study on student attitudes to AGI, wrote a Google Apps Script for the form, simulated 1,000 students, and analyzed the data. Result: From one short prompt, Manus produced a publishable-style report with charts from 1,000 synthetic responses. The data is made up. - Agents: Manus - Evidence: grade A (Shown, artifact linked), Author's blog with report excerpts, charts, form link and GitHub repo of files - Outcome: Completed - Job type: Research - Roles: Education - Industry: Universities - Setup: Manus agent with a central task-master, sub-agents and its own virtual computer. It built a to-do list and chose to write a Google Apps Script instead of driving Google Forms. - Caveat: It refused to submit fake responses to the real form on ethics grounds. All data is synthetic. - Prompt (written by ${BRAND}, not the author's exact prompt): I want to create a Google Form script for a research survey of [population] about [topic]. Mix Likert and open-ended questions, capture demographics. Then simulate [N] respondents, analyze the data, and write a publishable report. This is a proof of concept. - Evidence link: https://github.com/2020science/manus-synth-agi-survey-03-19-2025 - Source: https://futureofbeinghuman.com/p/when-agentic-ai-takes-charge-manus, 2025-03-22 - Published: 2026-10-13 ### Interactive course website from prompts URL: https://agenticfeats.com/c/interactive-course-website-from-prompts-manus/ Manus AI built a full AI-business course from one prompt, deployed it as a live website, then added quizzes, games, and embedded videos. Result: Three prompts produced a live interactive course website with modules, quizzes, and videos; the author also reports a 318-page ebook from further prompts. - Agents: Manus - Evidence: grade A (Shown, artifact linked), Live deployed course site linked; author describes outputs and shows a video - Outcome: Completed - Job type: Creative - Roles: Education, Founder or owner - Industry: Course creators - Setup: Manus AI, an autonomous agent, given plain-language prompts with no source material. It deployed the site to a manus.space URL with an in-page edit mode. - Caveat: The author says to verify the content, restrict it to trusted sources, and polish the initial design before publishing. - Prompt (written by ${BRAND}, not the author's exact prompt): Create an interactive course on [topic], covering [module list], with actionable tips, tools, and prompts. Then turn it into a website and deploy it permanently. Then add interactive quizzes and games to each module plus links to quality videos. - Evidence link: https://arortepv.manus.space/ - Source: https://aiblewmymind.substack.com/p/how-manus-ai-helped-me-create-a-318, 2025-06-12 - Published: 2026-10-13 ### Beginner programming book draft URL: https://agenticfeats.com/c/beginner-programming-book-draft-manus/ Manus wrote a beginner's book on programming with AI from a single request, using the free credits the author got to try it. Result: The author says the book came out well enough that they plan to publish it on Amazon. - Agents: Manus - Evidence: grade C (Claim, claim only), Author's own claim in a newsletter post; no book or screenshot shown - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner - Industry: Writers - Setup: Manus.ai on its free tier of 500 credits, given one prompt asking for a beginner's book on programming with AI. - Caveat: No book, page count, or sales figure is shown. Publication on Amazon is planned, not done. - Prompt (written by ${BRAND}, not the author's exact prompt): Write a book on programming with AI for complete beginners. Cover [topics], keep the tone [tone], and aim for about [length]. - Source: https://wmtips.substack.com/p/experimenting-with-ai-again, 2025-06-12 - Published: 2026-10-13 ### Strategy deck from one prompt URL: https://agenticfeats.com/c/strategy-deck-from-one-prompt-manus/ Manus Slides researched Generative AI 2025 trends from a vague one-line prompt and built a 16-slide consultancy-style PowerPoint deck. Result: It produced a 16-slide, research-backed deck with consistent design, published as a PDF; the author calls it the best AI slide builder they have used. - Agents: Manus - Evidence: grade A (Shown, artifact linked), Author links the finished 16-slide deck as a PDF download - Outcome: Partial - Job type: Marketing and Content - Roles: Founder or owner - Setup: Manus Slides given one vague prompt asking for a source-based strategic deck in UK English. No slide count or formatting guidance was supplied. - Caveat: Final slides had no source citations, page sizes varied, and some icons overlapped chart labels. These needed manual fixes in PowerPoint. - Prompt (written by ${BRAND}, not the author's exact prompt): Build a PPT presentation on [topic], focusing on [timeframe or angle]. Be accurate and source-based, cite all sources with a bibliography slide, use consultancy-grade design, and write in [language variant]. - Evidence link: https://buildingcreativemachines.substack.com/api/v1/file/6ef8c81d-4c5d-4148-bccb-467a14c237f7.pdf - Source: https://buildingcreativemachines.substack.com/p/manus-slides-my-experience-building, 2025-06-11 - Published: 2026-10-13 ### Manus tests on hotel booking engines URL: https://agenticfeats.com/c/manus-tests-on-hotel-booking-engines/ Manus.ai was prompted to book a September stay at hotels on 11 popular booking engines to test its navigation. Result: Manus succeeded on 8 of 11 booking engines, three of those with delays, and failed on D Edge, Simple Booking and Travelanium. - Agents: Manus - Evidence: grade B (Numbers, a specific number stated), Triptease article with per-engine results table; screenshots, no repo - Outcome: Partial - Job type: Research - Roles: Marketing, Operations - Industry: Hotels - Setup: Manus.ai, built on Anthropic's Claude and Alibaba's Qwen models, got the same one-line booking prompt for each hotel on each booking engine. - Caveat: Manus cannot store card or address details, so a human must finish at the payment page. Success meant reaching it. - Prompt (written by ${BRAND}, not the author's exact prompt): Please book me a room at [hotel name] from [check-in date] to [check-out date]. - Source: https://hotelnewsresource.com/article137200.html, 2025-07-09 - Published: 2026-10-13 ### Manus presentation test vs PageOn URL: https://agenticfeats.com/c/manus-presentation-test-vs-pageon/ Manus 1.6 Lite researched and built an 11-slide deck on AI trends in 2026, then was compared against PageOn.ai on the same prompt. Result: Manus took 10+ minutes and produced text-card slides with one chart; an add-charts edit took 7-8 minutes for three basic charts. - Agents: Manus - Evidence: grade A (Shown, artifact linked), Side-by-side slide screenshots and timings; author is a competitor's employee - Outcome: Partial - Job type: Marketing and Content - Roles: Marketing - Setup: Manus 1.6 Lite on the free tier, default settings, one prompt: AI Trends in 2026. A follow-up edit command asked for charts. - Caveat: Author works at PageOn.ai, the competing tool, so the comparison may be biased. Only one prompt was tested. - Prompt (written by ${BRAND}, not the author's exact prompt): Create a presentation on [topic]. After it is generated, add charts to wherever it needs. - Source: https://pageon.ai/blog/manus-ppt, 2026-02-12 - Published: 2026-10-13 ### Manus 15-task agent test URL: https://agenticfeats.com/c/manus-15-task-agent-test/ The JustPickAi editorial team gave Manus 15 real-world tasks across research, data, coding and content, then timed and scored each one. Result: In the published table, most tasks succeeded in 4 to 15 minutes, with 70% full success, 20% partial and 10% failure. - Agents: Manus - Evidence: grade B (Numbers, a specific number stated), Review with a per-task results table; no repo or demo linked - Outcome: Partial - Job type: Research - Setup: Manus run in its cloud-based virtual environment, with no extra tools or integrations stated. - Caveat: Technical tasks did poorly: Python debugging was partial and a multi-step API integration failed. The review also warns against using it for sensitive data. - Source: https://justpickai.com/blog/manus-ai-review-2026, 2026-02-20 - Published: 2026-10-13 ### Manus AI Meta ads test URL: https://agenticfeats.com/c/manus-ai-meta-ads-test/ Manus AI, run inside Meta Ads Manager on a live ecommerce client account, was asked to audit a performance drop and plan a fix. Result: It reported January ROAS as 14.25x when the real figure was 1.76x, and the author rated it unsafe for ecommerce scaling. - Agents: Manus - Evidence: grade A (Shown, artifact linked), Agency founder's review with screenshots; account data not independently shown - Outcome: Partial - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Marketing agencies - Setup: Manus AI embedded in Meta Ads Manager, given access to a live ecommerce client ad account with a 30-day versus January prompt. - Caveat: Useful for small lead gen accounts and reporting, but data must be double-checked and it only sees one ad account. - Prompt (written by ${BRAND}, not the author's exact prompt): Analyze the last 30 days versus [comparison month] for this ad account, explain why ROAS dropped, and give a plan to restore performance. - Source: https://skaleit.agency/blog/manus-ai-meta-ads-honest-review, 2026-04-30 - Published: 2026-10-13 ### Manus CRM and lead research URL: https://agenticfeats.com/c/manus-crm-and-lead-research/ Manus built a sponsorship-tracking CRM web app and researched trending YouTube topics and newsletter advertisers for a no-code education business owner. Result: A working kanban CRM with seven status stages and a 20-advertiser lead spreadsheet, with 3-4 hours of manual work done in 30-45 minutes. - Agents: Manus - Evidence: grade A (Shown, artifact linked), Sponsored blog post with screenshots and video; time savings stated by author - Outcome: Completed - Job type: Business - Roles: Founder or owner, Marketing - Industry: Course creators - Setup: Manus multi-agent tasks run in parallel, with its web app builder, browser research, and scheduled daily and weekly task automation. - Caveat: Some contact details were incomplete or misattributed, and the author says outputs need review. The post is sponsored by Manus. - Prompt (written by ${BRAND}, not the author's exact prompt): Research [newsletter niche] newsletters and give me [number] companies that advertised in them in the last [time period]. Include company name, newsletter source, ad copy, and contact info in a spreadsheet. - Evidence link: https://www.youtube.com/watch?v=3FESPBQ8hiI - Source: https://nocode.mba/articles/manus-ai-agents-automation, 2026-05-19 - Published: 2026-10-13 ### Cold outreach to car brands URL: https://agenticfeats.com/c/cold-outreach-to-car-brands-manus/ Manus searched for car brands without a web app, wrote an email, and sent it to them. Result: Some of the brands replied, and the author then built a web app for them. - Agents: Manus - Evidence: grade C (Claim, claim only), Author's own account with a photo; no numbers or links - Outcome: Completed - Job type: Sales - Roles: Founder or owner, Sales - Industry: Freelancers - Setup: Manus AI handled the search, email creation, and sending from a single request. - Caveat: The source gives no count of brands contacted or replies received, and no project details. - Prompt (written by ${BRAND}, not the author's exact prompt): Search for [type of business] that do not have a [web app or website]. Write a short email offering [your service] and send it to each one. - Source: https://x.com/jerryjenasom/status/2102306956203442534 by @JerryJenasom, 2026-09-22 - Published: 2026-10-13 ### Prague apartment hunt URL: https://agenticfeats.com/c/prague-apartment-hunt-manus/ Manus took one prompt with budget, district, and pet requirements, browsed Czech listing sites on its own, and compiled apartment options. Result: In about 5-6 minutes Manus returned five rental options with price, neighborhood, and links, work the author says would have taken hours. - Agents: Manus - Evidence: grade C (Claim, claim only), Author's first-hand account with screenshots; no listing output shared - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: Manus via invite code, running in the browser. One natural-language prompt, no listing sites named. It used Google.cz, Sreality, and Bezrealitky with browser translation. - Caveat: It mislabeled a 2+1 flat as two bedrooms, and looped when asked to email landlords, so the author took over. - Prompt (written by ${BRAND}, not the author's exact prompt): Find me a [number]-bedroom apartment for rent in [city]. Budget up to [amount] per month. Preferably in [districts], near a metro station. Needed from [date], [pet-friendly if needed]. Provide the best options with details and contacts. - Source: https://autogmail.com/manus-review - Published: 2026-10-13 ## Claude in Chrome ### Weekly ASDA grocery basket URL: https://agenticfeats.com/c/weekly-asda-grocery-basket-claude-in-chrome/ Claude in Chrome (Haiku) tried to fill an ASDA basket from a Claude-built Google Sheet of 40+ items, over five attempts. Result: Best run added 24 of 42 items for £71.76, then stopped early. The author still had to finish the basket and check out. - Agents: Claude in Chrome - Evidence: grade B (Numbers, a specific number stated), Author's detailed first-hand account with prices and item counts; screenshot only. - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: Claude in Chrome extension with Haiku, plus a Google Sheet Claude built with product names, quantities, frequency tags and price rules. Spreadsheet treated as the only source of truth. - Caveat: Slower than shopping by hand. It over-added from Regulars, hit token limits, gave up early, added unlisted carrots, and left checkout to the author. - Prompt (written by ${BRAND}, not the author's exact prompt): The spreadsheet is the ONLY source of truth for what to buy. You may use Regulars to find items faster, but only add items listed in the spreadsheet. Search using the exact product name from column [A]. Never add anything not listed. - Source: https://christinasouch.com/blog/can-claude-automate-my-grocery-shopping, 2026-01-27 - Published: 2026-10-13 ### Substack edits and migration URL: https://agenticfeats.com/c/substack-edits-and-migration-claude-in-chrome/ Claude in Chrome edited a published Substack post, migrated a blog post from a portfolio site, and added a note to an older post. Result: All three tasks finished and went live correctly, including a 324-offset insertion, 18+ link conversions, and no unwanted subscriber emails on edits. - Agents: Claude in Chrome - Evidence: grade B (Numbers, a specific number stated), Author's detailed first-person walkthrough with links to the live posts; no video. - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Newsletters - Setup: Claude in Chrome driving the Substack editor via screenshots, the accessibility tree, and JavaScript calls to the Tiptap/ProseMirror editor; link mapping table from a separate agent. - Caveat: The first insertion method merged text into the wrong paragraph and had to be undone before the second method worked. - Evidence link: https://priscasolutionsai.substack.com/p/my-people-i-owe-you-an-explanation - Source: https://priscasolutionsai.substack.com/p/i-walked-away-while-claude-edited, 2026-03-05 - Published: 2026-10-13 ### Browser tasks with Claude in Chrome URL: https://agenticfeats.com/c/browser-tasks-with-claude-in-chrome/ The author used Claude in Chrome to fill slide templates, structure meeting notes, compile research pages, and draft emails from the page on screen. Result: Slide building that took an hour now takes a fraction of that, and notes became formatted Word and PDF documents ready to share. - Agents: Claude in Chrome - Evidence: grade C (Claim, claim only), Author's own account; no output artifact or numbers shown - Outcome: Completed - Job type: Marketing and Content - Roles: Marketing - Setup: Claude in Chrome extension in the side panel on a paid Claude plan, with site permissions granted by the user. - Caveat: Slow on complex tasks, still in beta, and sometimes needs a nudge. Requires a paid Claude plan. - Prompt (written by ${BRAND}, not the author's exact prompt): Look at the page I have open and draft an outreach email to [recipient type] about [topic], pulling the key details from the page. - Source: https://marketsmarterwithai.substack.com/p/how-to-use-claudes-chrome-extension, 2026-03-24 - Published: 2026-10-13 ### Claude in Chrome PM review URL: https://agenticfeats.com/c/claude-in-chrome-pm-review/ Claude in Chrome was tested on help center docs, Google Ads campaign setup, and a Looker dashboard by a non-engineer product manager. Result: Docs rated 2 of 5, Google Ads setup 4 of 5, Looker dashboard 5 of 5; best docs run took about 2 hours. - Agents: Claude in Chrome - Evidence: grade A (Shown, artifact linked), Author's two-week review with screenshots; ratings are subjective, text is cut off - Outcome: Partial - Job type: Business - Roles: Marketing, Product - Industry: Saas - Setup: Claude in Chrome acting inside logged-in tools: Intercom, Google Docs, Google Ads, Looker. Haiku 4.5, Sonnet 4.5 and Opus 4.5 were compared on the docs task. - Caveat: Claude struggled with the Intercom editor, so docs went to Google Docs. Screenshot placement was awkward, and the best docs output was slow. - Source: https://rimo.app/en/blogs/claude-in-chrome-review1_en-US - Published: 2026-10-13 ### Meeting minutes into NotebookLM URL: https://agenticfeats.com/c/meeting-minutes-into-notebooklm-claude-in-chrome/ Claude in Chrome, on the Haiku model, extracted exact text of three years of W3C task force meeting minutes into markdown for NotebookLM. Result: Three years of minutes (2022-2024) were extracted in about 3 to 4 hours across three sessions and loaded as one source per year. - Agents: Claude in Chrome - Evidence: grade B (Numbers, a specific number stated), Author's first-person write-up with steps; no repo or output files shown - Outcome: Completed - Job type: Research - Roles: Operations - Setup: Claude in Chrome extension with the Haiku model, pointed at the minutes archive page. Output was requested as markdown in a code block, then copied into text files. - Caveat: It stalled after 3 or 4 months, so work was split into monthly chunks. A usage limit also hit once. Author spot-checked output manually. - Prompt (written by ${BRAND}, not the author's exact prompt): Go to each page of minutes for the meetings in [years]. From each page, extract the exact text of the minutes. Do not summarize. Present the text as a markdown file in a code block. - Source: https://lists.w3.org/Archives/Public/public-maturity/2026Feb/0004.html - Published: 2026-10-13 ## Custom agent ### Daily planning agent across work tools URL: https://agenticfeats.com/c/daily-planning-agent-across-work-tools-custom-agent/ The author built Plan My Today, a read-only agent and MCP server that pulls tasks from many work tools and prioritizes them for the day. Result: A dashboard and chatbot that merge tasks from calendar, mail, Linear, GitHub, Slack and meeting notes into one scored, prioritized view the author uses daily. - Agents: Custom agent - Evidence: grade A (Shown, artifact linked), Author's blog post with screenshots and a demo video - Outcome: Completed - Job type: Ops and IT - Roles: Developer - Setup: Custom MCP server with one tool per source, concurrent fetches, per-source priority scoring, a local JSON store, and Keycard token exchange for auth across seven provider resources. - Caveat: Agent is read-only by design; the author still changes task status in the source apps and tuned the scoring weights by hand. - Evidence link: https://www.youtube.com/watch?v=76xm1awRKOA - Source: https://dev.to/kimmaida/i-built-a-secure-planning-agent-with-mcp-and-keycard-324a, 2026-03-11 - Published: 2026-10-13 ### Agent researches its own competitors URL: https://agenticfeats.com/c/agent-researches-its-own-competitors-custom-agent/ Daemora, a self-hosted agent, searched Reddit, HN, Twitter and review sites unprompted and wrote a competitor analysis and go-to-market plan from one task. Result: It wrote a 3,000 word document with competitor pain points, 40 named target users, and an outbound playbook. - Agents: Custom agent - Evidence: grade B (Numbers, a specific number stated), Author's own account with figures; no output document linked - Outcome: Completed - Job type: Research - Roles: Founder or owner, Marketing - Setup: Daemora, the author's own self-hosted agent, using its webSearch, webFetch, browserAction, writeFile and createDocument tools. - Caveat: The author built the agent and is promoting it; the competitor claims and the 40 targets are not independently checked. - Prompt (written by ${BRAND}, not the author's exact prompt): You are [agent name]. Research the [market] landscape in [year]. Find the top competitors and analyse their pain points using real user complaints from Reddit, HN and Twitter. Identify target customers, write outreach scripts, and build a complete go-to-market strategy with cited sources. - Source: https://dev.to/dream_twister_4d65a3a36a4/i-asked-my-ai-agent-to-research-its-own-competitors-heres-what-it-found-4epd, 2026-05-13 - Published: 2026-10-13 ### Internal sales AI for Cursor reps URL: https://agenticfeats.com/c/internal-sales-ai-for-cursor-reps-custom-agent/ Cursor's Head of Enterprise Growth built ChatGTM, an internal chat tool that pulls CRM, call, and web data to draft outreach and research accounts. Result: Used across Cursor's 400+ person sales org, it reportedly helps SDRs book 3x more qualified meetings and cuts AE ramp time by over 50%. - Agents: Custom agent - Evidence: grade B (Numbers, a specific number stated), Newsletter interview with the builder; screenshots, no repo or independent data - Outcome: Completed - Job type: Sales - Roles: Operations, Sales - Industry: Software companies - Setup: Chat interface connected on demand to Salesforce, Gong, a data warehouse, LinkedIn, third-party data vendors, and the open web, with per-user and team memory. Native send via Gmail or Outreach. - Caveat: Results are reported by the author secondhand from George Hou; the article does not show the underlying measurements or methodology. - Source: https://thesignal.club/p/chatgtm, 2026-07-01 - Published: 2026-10-13 ### Pre-deploy release gatekeeper agent URL: https://agenticfeats.com/c/pre-deploy-release-gatekeeper-agent-custom-agent/ The author built SHIPCHECK, a ReAct agent that queries a Sanity Content Lake over MCP to check whether a service upgrade is safe before deploy. Result: On a payment service v2 to v4 test, it caught an outdated SDK, returned DO NOT SHIP YET, then SAFE TO SHIP in simulation. - Agents: Custom agent - Evidence: grade A (Shown, artifact linked), Repo, live demo site, and demo video; run trace shown in post - Outcome: Completed - Job type: Ops and IT - Roles: Developer, Operations - Setup: Custom ReAct agent using Sanity Content Lake with GROQ queries, exposed through four typed MCP tools. Documents encode components, version constraints, and authority-weighted knowledge entries. - Caveat: Demo runs against a purpose-built Sanity dataset; the source does not show it gating a real production deployment. - Prompt (written by ${BRAND}, not the author's exact prompt): Can I upgrade [service] from [current version] to [target version] in [environment] tonight? Check version constraints, live cluster dependencies, and conflicting docs, then give a ship or no-ship verdict with remediation steps. - Evidence link: https://github.com/aiwithrajan/shipcheck - Source: https://dev.to/rajan_mishra_a9f78ad216b4/shipcheck-an-autonomous-react-agent-that-stops-cloud-outages-before-they-happen-5ag1, 2026-09-26 - Published: 2026-10-13 ### Rule-citing AI dungeon master game URL: https://agenticfeats.com/c/rule-citing-ai-dungeon-master-game-custom-agent/ The author built a browser D&D 5e dungeon crawl whose AI Dungeon Master answers rules questions by looking up and citing rules stored in Sanity. Result: A live game with 87 rules, 33 conditions, 14 spells and 15 monsters in two editions, answering rules questions in about 1.3 to 2.2 seconds. - Agents: Custom agent - Evidence: grade A (Shown, artifact linked), Live demo, GitHub repo, screenshots and GIFs - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Sanity dataset seeded with SRD data, two Context MCP endpoints (knowledge base and GROQ), DeepSeek on Baseten via AI SDK. Server prefetches lookups before the model runs. - Caveat: The MCP endpoints need an org token with the Context Viewer role. A project token returns a 403 error, which cost the author an hour. - Evidence link: https://d-and-d-one.vercel.app/ - Source: https://dev.to/prime_299792/the-goblin-warren-an-ai-dungeon-master-that-never-invents-a-rule-4d66, 2026-10-04 - Published: 2026-10-13 ### Agent-orchestrated AI video URL: https://agenticfeats.com/c/agent-orchestrated-ai-video-custom-agent/ The author's custom agent dimul used Grok through the CLI as an orchestrator, generating clips with Imagine and assembling them into a finished video. Result: The orchestrator produced a complete video with clips, music, and effects, shown in a linked video. - Agents: Custom agent, Grok Bot - Evidence: grade A (Shown, artifact linked), Linked video of the output; no repo or numbers - Outcome: Completed - Job type: Creative - Roles: Developer - Setup: Custom agent dimul connected to Grok via the CLI on a SuperGrok subscription. Grok 4.7 orchestrates Imagine for clip generation, then handles assembly, music, and effects. - Caveat: The agent is not yet released; the author says final touches remain. No time or cost figures are given. - Evidence link: https://x.com/nurm_dima/status/2107228763138245055/video/1?s=46 - Source: https://x.com/xfreeze/status/2107878966510158087 by @NURM_Dima, 2026-10-07 - Published: 2026-10-13 ### Self-hosted Matrix agent assistant URL: https://agenticfeats.com/c/self-hosted-matrix-agent-assistant-custom-agent/ The author had Kimi set up a local Matrix server and write a small custom agent loop with sandboxed tools. Result: A working personal agent the author reaches by Matrix from phone, home web app, or CLI agents. - Agents: Custom agent - Evidence: grade C (Claim, claim only), Author describes their own setup in detail; no repo or demo shared - Outcome: Completed - Job type: Setup and Meta - Roles: Personal - Setup: Kimi and GLM5 run the agents. Tools run under bubblewrap sandboxing. A self-hosted Matrix server with federation off, behind WireGuard, plus identity, soul, and memory files, a heartbeat, and reminders. - Caveat: Bubblewrap is not enough for untrusted data, so a VM is planned. Agents have web access, so leaks are possible. - Source: https://news.ycombinator.com/item?id=50003674 by vidarh, 2026-10-08 - Published: 2026-10-13 ### Hospital-style agent pipeline ships Db2 support URL: https://agenticfeats.com/c/hospital-style-agent-pipeline-ships-db2-support-custom-agent/ Cockroach Labs ran a multi-role coding agent pipeline modeled on a teaching hospital to add Db2 support to its MOLT migration tool. Result: Db2 support shipped in under two days for a $4172 token bill, versus 9 months and about $160K for Oracle; 27 PRs merged. - Agents: Custom agent - Evidence: grade A (Shown, artifact linked), Company blog with figures, issue screenshot and Sinai repo PR link - Outcome: Completed - Job type: Coding - Roles: Developer - Industry: Software companies - Setup: Agents run on GitHub Actions with roles for triage, workup, implementation, review, and discharge. Issue labels track each patient's state, and a human is consulted only on escalation. - Caveat: The review structure adds bureaucracy, wait times, and cost compared with a swarm or single agent; the authors accept this for quality. - Evidence link: https://github.com/cockroachlabs/sinai/pull/691 - Source: https://cockroachlabs.com/blog/experiment-running-hospital-code, 2026-10-09 - Published: 2026-10-13 ## Replit Agent ### Podcast website built with Replit URL: https://agenticfeats.com/c/podcast-website-built-with-replit-replit-agent/ The author used Replit to generate a clickable website prototype for their podcast and newsletter, then finished the MVP with help from Val Town. Result: A live MVP website for the podcast and newsletter went up after about 10 hours of struggling to load it on a custom URL. - Agents: Replit Agent - Evidence: grade A (Shown, artifact linked), Author describes the build and links the live site - Outcome: Completed - Job type: Coding - Roles: Founder or owner - Industry: Podcasters - Setup: Replit generated the prototype. Val Town staff pair programmed to get it onto the custom domain. Canva, Riverside, and Descript handled the logo, trailer, and editing. - Caveat: Over 10 hours went to failing to deploy to a custom URL in Replit before outside help fixed it. - Evidence link: https://www.thenextnext.ai/ - Source: https://thenextnext.substack.com/p/the-next-next-update-5, 2025-01-09 - Published: 2026-10-13 ### Road safety reporting web app URL: https://agenticfeats.com/c/road-safety-reporting-web-app-replit-agent/ Replit Agent built a full web app where the public reports bad driving, including a Postgres database, image uploads, Firebase login, and a landing page. Result: The app went live on a custom domain, dashdiary.co.uk, for a total build cost of $35.82 plus a few dollars for the domain. - Agents: Replit Agent - Evidence: grade A (Shown, artifact linked), Live site linked, with a line-by-line cost breakdown from the author - Outcome: Completed - Job type: Coding - Roles: Developer, Founder or owner - Setup: Replit Agent and Assistant on a $25 monthly plan. Postgres on Neon, UploadThing for images, Firebase for sign-up, Google Analytics, Namecheap domain, Mailerlite welcome email. - Caveat: Swapping in Firebase auth broke some features and needed fixing. The first deploy failed, and the dev and production databases are the same instance. - Evidence link: https://dashdiary.co.uk/ - Source: https://thedatacoach.substack.com/p/how-i-used-replit-to-build-a-fully, 2025-01-20 - Published: 2026-10-13 ### Newsletter summariser app URL: https://agenticfeats.com/c/newsletter-summariser-app-replit-agent/ Replit Agent built a Gmail newsletter summariser for a low-coder, with batch summaries, deduplication, archiving and export, over three days. Result: The author shipped a working summariser targeting a cut from 2-3 weekly hours of newsletter reading to 30 minutes. - Agents: Replit Agent - Evidence: grade A (Shown, artifact linked), Author's account with screenshots; no repo, app shared on request - Outcome: Completed - Job type: Coding - Roles: Developer, Founder or owner - Setup: Replit Agent, driven by natural language prompts, connected to Gmail. The author has basic Python knowledge. - Caveat: Needs a Google email account, and the author notes clear limitations of the tool. Time saved is the target, not a measured result. - Prompt (written by ${BRAND}, not the author's exact prompt): Develop a simple [app type] with a clean, responsive interface and [number] main functions: [function 1], [function 2], [function 3]. Save data locally so it persists between sessions. - Source: https://justinyjtan.substack.com/p/building-with-replit-agent-low-coders-field-notes, 2025-01-07 - Published: 2026-10-13 ### Teleprompter Chrome extension published URL: https://agenticfeats.com/c/teleprompter-chrome-extension-published-replit-agent/ Replit Agent built a floating teleprompter overlay extension for browser video meetings, refined with Claude and ChatGPT. Result: The extension was accepted and published on the Chrome Web Store, and the author says it solves their own habit of looking down at notes. - Agents: Replit Agent - Evidence: grade A (Shown, artifact linked), Live Chrome Web Store listing linked by the author - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Replit did about 80-85% of the build. Claude and ChatGPT were used for further refinement of the extension. - Caveat: No usage numbers or reviews are given, and the author did not share the prompts they used. - Evidence link: https://chromewebstore.google.com/detail/teleprompter-overlay/hbplgdlkkejlpiakenbjpmgolcbnakfn - Source: https://replit.discourse.group/t/chrome-extension-got-published-yay/4845, 2025-06-05 - Published: 2026-10-13 ### Tattoo booking app prototype URL: https://agenticfeats.com/c/tattoo-booking-app-prototype-replit-agent/ Replit Agent built a booking and payment app for tattoo artists from prompts, covering calendar, Google OAuth, Stripe, and Postmark emails, over one weekend. Result: The author, a non-developer for a decade, built a working end-to-end booking app with artist dashboards in 48 hours. It is not yet launched. - Agents: Replit Agent - Evidence: grade C (Claim, claim only), Author's first-person account; no repo or demo linked - Outcome: Completed - Job type: Coding - Roles: Founder or owner, Product - Industry: Tattoo studios - Setup: Replit Agent in a blank repl, prompted with a Calendly-like calendar, Google OAuth, Stripe deposits, and Postmark notifications. React front end. - Caveat: The agent sometimes rewrote working code and claimed it fixed things it had broken. Stripe payments are on the roadmap, and the app is unlaunched. - Prompt (written by ${BRAND}, not the author's exact prompt): Build a booking app like Calendly for [profession]. Include Google OAuth, Stripe deposits, and Postmark email confirmations. Clients see availability and book; [professionals] get a dashboard. - Source: https://iliashatzis.substack.com/p/from-zero-to-working-prototype-in, 2025-07-07 - Published: 2026-10-13 ### Automated podcast website URL: https://agenticfeats.com/c/automated-podcast-website-replit-agent/ Replit Agent built and deployed a podcast site that imports episodes from RSS and generates transcripts, tags, and companion content automatically. Result: The author built a live podcast site in one weekend that creates companion resources when a show is published; AWS Transcribe cut transcript costs 73%. - Agents: Replit Agent - Evidence: grade A (Shown, artifact linked), Live site linked; author states build time; feature list comes from the agent's own briefing - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Podcasters - Setup: Replit Agent with React, Express, PostgreSQL, AWS Transcribe, SES, S3, OpenAI GPT-4o and GPT-4o-mini, Firebase auth, and Beehiiv newsletter signup. - Caveat: Took a weekend with lots of back and forth to get the result the author wanted. - Evidence link: https://podcast.thekamclub.com/ - Source: https://replit.discourse.group/t/how-to-create-an-automated-podcast-website-with-replit/6034, 2025-07-22 - Published: 2026-10-13 ### Micro-SaaS built and deployed URL: https://agenticfeats.com/c/micro-saas-built-and-deployed-replit-agent/ The author used Replit Agent to build and deploy Calendar Autobot, an app that turns forwarded emails and images into Google Calendar events. Result: A production micro-SaaS went live in about 3 hours, versus an estimated 3 to 4 weeks solo. - Agents: Replit Agent - Evidence: grade A (Shown, artifact linked), Live app and Chrome extension linked; time claim self-reported - Outcome: Completed - Job type: Coding - Roles: Developer, Founder or owner - Setup: Replit paid plan with Agent. Stack: Python, Flask, PostgreSQL, OpenAI, Google OAuth, Mailgun. Starter prompt generated in Google AI Studio. Early logging helped the agent debug. - Caveat: The agent bloated code and looped on Google OAuth and Mailgun. The author reviewed code and wrote specific prompts with docs links to steer it. - Prompt (written by ${BRAND}, not the author's exact prompt): Using Python Flask, create a backend API that takes [email or document text], extracts every event (name, description, start and end date and time, location) as a JSON list, resolves relative dates from [CURRENT DATE], and creates matching Google Calendar events. - Evidence link: https://calautobot.com/ - Source: https://dev.to/willcheung/i-built-and-deployed-a-micro-saas-in-3-hours-heres-what-i-learned-cd6, 2025-08-31 - Published: 2026-10-13 ### Interactive map of funded companies URL: https://agenticfeats.com/c/interactive-map-of-funded-companies-replit-agent/ Replit Agent built and hosted a searchable map of funded brick-and-mortar companies on Wefunder from the author's prompts, using the Leaflet map library. Result: About 3 hours and roughly $22 produced a live, searchable map of funded companies, versus an estimated 4 to 5 hours by hand. - Agents: Replit Agent - Evidence: grade A (Shown, artifact linked), Live demo and Replit repo linked; author states time and cost - Outcome: Partial - Job type: Coding - Roles: Developer - Setup: Replit Agent with Leaflet for the map. The author supplied a private Wefunder endpoint, then parsed the explore page JSON themselves and hardcoded the data. - Caveat: The agent wasted credits debugging the API and reloading the app. Small UI edits cost $0.30 to $2 each. Search still needs work. - Evidence link: https://wefunder-brick-and-mortar-map.replit.app/ - Source: https://dev.to/kenakingkong/building-an-interactive-map-with-replit-agent-a-software-engineer-tests-vibe-coding-efficiency-3068, 2025-11-17 - Published: 2026-10-13 ### Invoice processing app on Replit URL: https://agenticfeats.com/c/invoice-processing-app-on-replit-replit-agent/ The author built Documind, an invoice platform, on Replit by prompting Claude, with extraction, anomaly detection, workflows, and plain-English queries. Result: Documind extracts invoice data in under four seconds; a 130-invoice demo caught four anomalies, including an $875 duplicate and a 60% vendor price increase. - Agents: Replit Agent - Evidence: grade A (Shown, artifact linked), Live Replit app link and demo figures; no independent verification - Outcome: Completed - Job type: Finance and Admin - Roles: Finance, Founder or owner - Setup: Replit as the dev environment, prompting for the build. React, Vite, Node.js, Express, SQLite, and the Claude API for extraction and natural language queries. - Caveat: Tested only on simulated invoices for a fictional coffee shop; the $7,800 yearly saving is the author's own estimate, not a measured result. - Evidence link: https://cc921e05-8cd3-415f-bee6-5467beb99843.replit.app/ - Source: https://medium.com/@nthamma1/i-built-an-ai-powered-invoice-platform-on-replit-using-prompts-heres-what-happened-44e35c3d5c31, 2026-03-06 - Published: 2026-10-13 ### Three products in one day URL: https://agenticfeats.com/c/three-products-in-one-day-replit-agent/ Replit Agent rebuilt a full-stack platform, added seven features to a live RAG app, and produced a demo video during a free Agent day. Result: About 12 hours produced 46 commits and over 14,000 lines of code across three shipped products. - Agents: Replit Agent - Evidence: grade B (Numbers, a specific number stated), Author states commit and line counts; photo attached, no repo link - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Replit Agent, used during a free Agent day offered by Replit. - Caveat: Relies on a free Agent day promotion; no repo or live links given, only a photo attached. - Source: https://x.com/lakshonline/status/2050977394861990332 by @LakshOnline, 2026-05-03 - Published: 2026-10-13 ### GiveABook book sharing app URL: https://agenticfeats.com/c/giveabook-book-sharing-app-replit-agent/ Replit Agent built GiveABook, an app where people list books they have read, drop them off, and neighbors pick them up on Sundays. Result: The author shipped a live book-sharing app and its demo video, both made with Replit Agent, during a 24-hour buildathon. - Agents: Replit Agent - Evidence: grade A (Shown, artifact linked), Live app link and demo video; no usage numbers - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Replit Agent used inside Replit for the full app and the demo video. - Caveat: The source gives no usage numbers or details on manual edits, and it is a hackathon entry. - Evidence link: https://zengy-reads-premium--thippeshdigital.replit.app - Source: https://x.com/athippesh08/status/2050931383846273333 by @AThippesh08, 2026-05-03 - Published: 2026-10-13 ## ChatGPT agent ### UNESCO education data spreadsheet URL: https://agenticfeats.com/c/unesco-education-data-spreadsheet-chatgpt-agent/ ChatGPT Agent pulled UNESCO student-teacher ratio and teaching staff data and built a multi-tab spreadsheet in a single 14-minute run. Result: The Agent worked for 14 minutes on the UNESCO task; the author tested it against o3-pro and ran four other work-style tasks. - Agents: ChatGPT agent - Evidence: grade A (Shown, artifact linked), Tutorial with screenshots and a shared ChatGPT conversation link; text excerpt cuts off before outputs. - Outcome: Partial - Job type: Research - Roles: Education - Setup: ChatGPT Agent on a paid plan, using its virtual computer with browser, terminal, code execution, and file system. - Caveat: The author reports weaker results on the image collage and Lego parts tasks, which needed manual fixes or only partly worked. - Prompt (written by ${BRAND}, not the author's exact prompt): Pull the latest data from [data source] on [metric] for [groups]. Build a spreadsheet with one tab per [group], listing [columns]. Add a summary tab with the best and worst [metric] values, sorted. Keep [naming field] consistent across tabs. - Evidence link: https://chatgpt.com/s/6880dcdff5b08191b3736d97abd1056f - Source: https://datacamp.com/tutorial/chatgpt-agent, 2025-07-24 - Published: 2026-10-13 ### Booking tests with ChatGPT Agent URL: https://agenticfeats.com/c/booking-tests-with-chatgpt-agent/ ChatGPT Agent was tested on restaurant and software demo bookings, with mixed results across three attempts. Result: The VICE booking took over eight minutes and four tries on the time picker, while the SE Ranking demo booking took 37 seconds. - Agents: ChatGPT agent - Evidence: grade B (Numbers, a specific number stated), Author's screenshots and timings; no repo - Outcome: Partial - Job type: Research - Roles: Marketing - Setup: ChatGPT Agent run in its browser mode, given plain-language booking requests for a restaurant and a software demo. - Caveat: It could not call the restaurant, asked for the missing phone number only after eight minutes of workarounds, and the gated-form test is cut off. - Source: https://seranking.com/blog/chatgpt-agent-digital-marketing-review, 2025-07-24 - Published: 2026-10-13 ### Restaurant list and home search URL: https://agenticfeats.com/c/restaurant-list-and-home-search-chatgpt-agent/ ChatGPT agent picked 10 restaurants in the author's vacation city, found signature dishes, and saved them as a Google Maps list with notes. Result: Built the 10-restaurant Maps list in 23 minutes and exported 13 Marbella home listings to Excel, with data the author checked as accurate. - Agents: ChatGPT agent - Evidence: grade A (Shown, artifact linked), Author's screen recordings and spreadsheet screenshot; author verified the data against the site - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: ChatGPT agent in browser mode, with Google Maps access after the author logged in manually, and a real estate site for the home search. - Caveat: It needed human help for login and a stuck price dropdown, took about 20 minutes, and each interruption uses one monthly message credit. - Prompt (written by ${BRAND}, not the author's exact prompt): Find the best restaurants in [city], pick [number], note the signature dish at each, and save them as a new list on Google Maps with those notes. - Source: https://artificialcorner.com/p/my-honest-review-of-chatgpt-agent, 2025-07-18 - Published: 2026-10-13 ### Annotated bibliography on big data URL: https://agenticfeats.com/c/annotated-bibliography-on-big-data-chatgpt-agent/ ChatGPT Agent searched academic databases, filtered peer-reviewed open-access papers, and compiled an annotated bibliography on big data in education and healthcare decision-making. Result: It produced a review-ready draft with citations and a gaps synthesis in under ten minutes, versus half a day by hand. - Agents: ChatGPT agent - Evidence: grade A (Shown, artifact linked), Author states time saved and links a downloadable PDF of the output - Outcome: Completed - Job type: Research - Roles: Education - Setup: ChatGPT Agent run with the /agent command, searching PubMed, ScienceDirect, ResearchGate, Springer, and PMC. Output exported to markdown, then converted to a .docx. - Caveat: Sources must be verified, interpretations need human review, and formatting sometimes needs cleanup. - Prompt (written by ${BRAND}, not the author's exact prompt): /agent Find recent peer-reviewed papers (past five years) on [topic], especially in [field 1] and [field 2]. Compile an annotated bibliography and summarise the key research gaps. - Evidence link: https://www.dropbox.com/scl/fi/dq4ntyovy37jnnhg45nmo/Big-Data-in-decision-making.pdf?rlkey=ldwyo86m4a5k0invmzgb4n0ua&st=rt1kjecz&dl=0 - Source: https://datawithsultan.medium.com/i-put-chatgpt-agent-to-the-test-heres-how-it-helped-me-write-an-academic-paper-984327423d95, 2025-07-28 - Published: 2026-10-13 ### Competitor analysis deck and table URL: https://agenticfeats.com/c/competitor-analysis-deck-and-table-chatgpt-agent/ ChatGPT Agent mode researched the top 5 UK coffee subscription companies and built a competitor deck plus a spreadsheet of each company's top 5 USPs. Result: One prompt returned a PDF deck and a USP spreadsheet in 18 minutes, with positioning, pricing models and a SWOT section. - Agents: ChatGPT agent - Evidence: grade A (Shown, artifact linked), Author shows screenshots of the deck and spreadsheet; states 18-minute run time - Outcome: Completed - Job type: Research - Roles: Marketing - Setup: ChatGPT Pro with Agent mode enabled. One specific prompt naming industry, geography and the two export formats. Optionally run the output through Gamma to polish the deck. - Caveat: Output is a good first pass; the author keeps prompting to refine it and suggests a tool like Gamma to make it presentable. - Prompt (written by ${BRAND}, not the author's exact prompt): Create a competitor analysis deck for the top 5 [country] [industry] companies and export it as a PDF. Also create a table with each company's top 5 USPs and export it as a spreadsheet. - Source: https://marketsmarterwithai.substack.com/p/how-to-run-a-full-competitor-analysis, 2025-08-05 - Published: 2026-10-13 ### Presentation and eBook from prompts URL: https://agenticfeats.com/c/presentation-and-ebook-from-prompts-chatgpt-agent/ ChatGPT Agent researched AI in education and built a branded slide deck, and merged past newsletter posts with approved sources into an eBook draft. Result: Produced a functional researched presentation with the author's logo and colors, plus a PDF eBook draft; design was plain, so both are drafts. - Agents: ChatGPT agent - Evidence: grade A (Shown, artifact linked), Linked Google Slides deck and Drive file; author describes results - Outcome: Partial - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Newsletters - Setup: ChatGPT Agent Mode with browsing, Google Drive file access, and a detailed prompt restricting external sources to OpenAI, Anthropic, and Google docs. - Caveat: Slides were plain and tasks run slowly. The author says output needs fact-checking and a detailed prompt to be usable. - Prompt (written by ${BRAND}, not the author's exact prompt): You are my research assistant and editor. Read my uploaded [documents], cluster them into themes, find gaps, fill them using only [approved sources], draft a [page count]-page outline, expand it into a full draft, and format it as a PDF with title page, intro, chapters, conclusion, and references. - Evidence link: https://docs.google.com/presentation/d/1SwKVhTwZmEI579PT470XrC-4kUehE5bP/edit?usp=drive_link&ouid=105756463132153665761&rtpof=true&sd=true - Source: https://aiblewmymind.substack.com/p/i-tested-chatgpt-agent-on-real-work, 2025-08-28 - Published: 2026-10-13 ### Buying on Amazon by photo URL: https://agenticfeats.com/c/buying-on-amazon-by-photo-chatgpt-agent/ The author used ChatGPT Agent to buy items on Amazon from a photo of the old item, handling the full purchase. Result: The agent completed whole Amazon purchases from a wrapper photo and a short instruction, until Amazon blocked agent access. - Agents: ChatGPT agent - Evidence: grade C (Claim, claim only), Author's own account in a comment; no screenshot or link - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: ChatGPT Agent with a photo of the item and a plain instruction to buy a replacement on Amazon. - Caveat: Amazon later blocked agent purchases, so this no longer works. The author moved some shopping to other retailers. - Prompt (written by ${BRAND}, not the author's exact prompt): Here is a photo of [item or packaging]. Buy a replacement on [retailer] and complete the checkout. - Source: https://news.ycombinator.com/item?id=45815103 by pants2, 2025-11-04 - Published: 2026-10-13 ### AI browser shopping test URL: https://agenticfeats.com/c/ai-browser-shopping-test-perplexity-comet/ DreamHost tested five AI browsers on one Amazon task: research grinder coffee makers under $100, tabulate five, add two to cart. Result: Perplexity Comet completed the task and stopped at the cart. ChatGPT Atlas also reached checkout but was too slow to be usable. - Agents: ChatGPT agent, Other, Perplexity Comet - Evidence: grade A (Shown, artifact linked), Blog write-up with screenshots; the author states timing observations, no repo. - Outcome: Partial - Job type: Research - Roles: Marketing - Setup: Perplexity Comet, ChatGPT Atlas Agent mode, Opera Neon, Dia and Chrome with Gemini, each given the same natural language prompt on Amazon. - Caveat: The excerpt is cut off before the Opera Neon, Dia and Chrome results. Testing was one shopping task plus research prompts. - Prompt (written by ${BRAND}, not the author's exact prompt): Go to [shopping site], research the best [product type] under [price] with [feature], and create a table of the 5 best. Then add two of the best choices to cart with my [city] address selected. - Source: https://dreamhost.com/blog/ai-web-browser, 2026-01-05 - Published: 2026-10-13 ### Matcha ordered by agent URL: https://agenticfeats.com/c/matcha-ordered-by-agent-claude-cowork/ The author connected Monid API keys to Claude Cowork, which customized the milk and ordered a coconut cloud matcha at a tech week pop-up. Result: A coconut cloud matcha was ordered by the agent at the first SF tech week event. The author also had ChatGPT sign up for events. - Agents: ChatGPT agent, Claude Cowork - Evidence: grade A (Shown, artifact linked), Author's post with photos and a video; no repo - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Monid API keys connected to Claude Cowork. A separate ChatGPT setup picked events from what it knows about the author. - Caveat: A small personal errand. The event sign-ups are mentioned in one line with no count. - Source: https://x.com/gracepace_/status/2107346389282021501 by @gracepace_, 2026-10-06 - Published: 2026-10-13 ### Multi-agent stock analysis pipeline URL: https://agenticfeats.com/c/multi-agent-stock-analysis-pipeline-cursor/ The author built a LangGraph system where valuation, sentiment and fundamental agents plus a coordinator rate four stocks and backtest the resulting portfolio. Result: A working pipeline that outputs picks, performance metrics and a chart, with a 63-day backtest against an equal-weight benchmark and a test suite. - Agents: ChatGPT agent, Cursor, Custom agent - Evidence: grade A (Shown, artifact linked), Repo with README, tests and described output files; no results shown - Outcome: Completed - Job type: Finance and Admin - Roles: Developer, Finance - Setup: Cursor and Claude for code and prompts, ChatGPT Agent mode for news and SEC filing data, LangGraph for orchestration, financialdatasets.ai for prices, VADER for sentiment. - Caveat: Prototype with no transaction costs, one backtest window (Aug to Nov 2024), survivorship bias, and thresholds chosen by trial rather than tuned. - Source: https://github.com/phanhpp/Multi-Agent-Stock-Analysis-System - Published: 2026-10-13 ## Hermes Agent ### AI-made 8D music video URL: https://agenticfeats.com/c/ai-made-8d-music-video-hermes-agent/ Hermes-Agent wrote lyrics, generated music, vocals and stems, made images from the lyrics, and added audio-reactive effects to produce a full music video. Result: The agent produced a complete 8D music video with seven locally rendered images timed to the lyric structure. - Agents: Hermes Agent - Evidence: grade C (Claim, claim only), Author describes the workflow; the video is in a linked earlier post. - Outcome: Completed - Job type: Creative - Roles: Personal - Setup: Hermes-Agent was asked to give itself skills. It used ACE-Step-1.5 for music, vocals and stems, and Flux-Klein-4B (flux2.c) for images, all run locally. - Caveat: Only seven images were rendered because of local tools. The author made the two character images by hand and wants to add transcribed lyrics later. - Prompt (written by ${BRAND}, not the author's exact prompt): Give yourself the skills to make a music video. Write lyrics about [theme], generate music with vocals and stems, create image prompts from the lyrics, render the images, time them to the lyric structure, and add audio-reactive effects. - Evidence link: https://x.com/alexcovo_eth/status/2034310341585174878 - Source: https://x.com/alexcovo_eth/status/2034804392507912392 by @alexcovo_eth, 2026-03-20 - Published: 2026-10-13 ### Weekly competitive briefing skill URL: https://agenticfeats.com/c/weekly-competitive-briefing-skill-hermes-agent/ Hermes Agent ran the same Monday competitive monitoring prompt for six weeks and rewrote its own underlying skill four times. Result: The same task fell from 20 minutes in week one to 12 in week four and 8 by week six, with better competitor patterns surfaced. - Agents: Hermes Agent - Evidence: grade B (Numbers, a specific number stated), Author states timings from their logs; chart image, no repo - Outcome: Completed - Job type: Research - Roles: Product - Setup: Hermes Agent from Nous Research, with skills saved as files in ~/.hermes/skills/. It can run on Claude, GPT-4o, Gemini, or local Llama models. - Caveat: Full skill files and the rollout plan are behind a paywall, and the timings are the author's own logs. - Source: https://news.aakashg.com/p/hermes-agent-guide, 2026-05-01 - Published: 2026-10-13 ### Voice notes into Obsidian notes URL: https://agenticfeats.com/c/voice-notes-into-obsidian-notes-hermes-agent/ Hermes, run as an agent named Satori, turns the author's Telegram voice notes into structured Obsidian notes for business ideas and newsletter drafts. Result: A 500-bullet idea list became researched one-page notes with metadata, and a dog-walk voice dump became a shaped newsletter draft. - Agents: Hermes Agent - Evidence: grade A (Shown, artifact linked), Author's detailed account with screenshots; no numbers on time saved - Outcome: Completed - Job type: Business - Roles: Founder or owner - Industry: Writers - Setup: Hermes agent on a Mac Mini, driven through Telegram, writing into a local Obsidian vault. Can also hand specs to Claude Code or Codex. - Caveat: The author calls the system messy and says most people don't need it in its current shape. - Prompt (written by ${BRAND}, not the author's exact prompt): Transcribe my voice note about [idea]. Create a note in [vault folder] with metadata and tags. Keep the verbatim transcript, add a cleaned version, then research competitors, open questions, differentiation, and MVP scope. - Source: https://metedata.substack.com/p/013-my-hermes-and-obsidian-set-up, 2026-05-20 - Published: 2026-10-13 ### Travel agent that learns from trips URL: https://agenticfeats.com/c/travel-agent-that-learns-from-trips-hermes-agent/ Hermes Agent plans trips using stored traveler preferences, researches flights and hotels by web search, and rewrites its own memory after a post-trip debrief. Result: After a debrief on a 6-hour layover, the agent added a 3-hour layover cap and applied it unprompted when re-planning the next trip. - Agents: Hermes Agent - Evidence: grade A (Shown, artifact linked), Public repo and a recorded demo walkthrough; data was seeded - Outcome: Completed - Job type: Personal and Family - Roles: Operations - Industry: Travel agencies - Setup: Hermes Agent with two custom skills (trip planning, post-trip debrief), a context file, MEMORY.md and SOUL.md. Runs on gemini-3-flash-preview with built-in web search and memory. - Caveat: Traveler profiles and prior-trip data were seeded before the demo, and trip dates are fixture data, not live calendar events. - Prompt (written by ${BRAND}, not the author's exact prompt): Plan a trip for [traveler] to [destination] [timeframe]. Apply all stored preferences, research flights and hotels, check visa validity, and end with a pending-approval summary. - Evidence link: https://github.com/tourswithchris/hermes-travel-ops-native - Source: https://dev.to/christopher_mburu_7b7e0ba/hermes-travelops-an-agent-that-learns-from-every-trip-5805, 2026-05-31 - Published: 2026-10-13 ### SaaS accounting automation URL: https://agenticfeats.com/c/saas-accounting-automation-hermes-agent/ Hermes Agent, connected to Zoho Books, Gmail, Aspire Bank, Stripe and Telegram, ran bookkeeping for the author's Singapore SaaS company. Result: Setup took about 6 to 8 hours of active work and cut routine month-end bookkeeping by an estimated 85 to 90 percent. - Agents: Hermes Agent - Evidence: grade B (Numbers, a specific number stated), Author's own account with workflow and error list; reduction is an estimate - Outcome: Completed - Job type: Finance and Admin - Roles: Finance, Founder or owner - Industry: Saas - Setup: Hermes Agent with Zoho Books, Gmail, Aspire Bank, Stripe and Telegram. No MCP servers; the agent wrote its own Python scripts against the official APIs. - Caveat: The author corrected seven errors, including a critical tax-amount bug on vendor bills. It runs in parallel with a human accountant. - Source: https://satvasolutions.com/blog/how-we-automated-accounting-ai-agent, 2026-06-02 - Published: 2026-10-13 ### Home server chat for teen URL: https://agenticfeats.com/c/home-server-chat-for-teen-hermes-agent/ A Hermes agent running on a local Qwen3.6-35B model set up an Open WebUI server and handled account management from the author's phone. Result: The author's teen now has their own account on a home-network local model for chat, research and troubleshooting, with no outside provider needed. - Agents: Hermes Agent - Evidence: grade C (Claim, claim only), Author's own account in a comment; no repo or screenshots - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Qwen3.6-35B-A3B served through llama.cpp on a home server, Open WebUI as the interface, and Hermes Agent driven from a phone. - Caveat: The author still keeps a personal Anthropic subscription and says SoTA models make sense for sophisticated coding agent work. - Source: https://news.ycombinator.com/item?id=48650822 by roadside_picnic, 2026-06-23 - Published: 2026-10-13 ### Hermes as personal operating layer URL: https://agenticfeats.com/c/hermes-as-personal-operating-layer-hermes-agent/ Hermes Agent grew from a chat interface into an operating layer across the author's work, research, personal organisation and family life over fifteen days. Result: The author added over 50 capabilities in 15 days, including email triage, daily briefings and a custom MCP server for tasks and invoices. - Agents: Hermes Agent - Evidence: grade A (Shown, artifact linked), Day-by-day timeline site; author states 50+ capabilities and 15 days - Outcome: Completed - Job type: Setup and Meta - Roles: Founder or owner, Personal - Setup: Hermes Agent with a custom MCP server connecting it to the author's own system for tasks, projects, clients and invoices. Local models and many added capabilities. - Evidence link: https://fifteen-days-hermes.pages.dev/ - Source: https://x.com/marc_bara/status/2099052224496140593 by @marc_bara, 2026-09-13 - Published: 2026-10-13 ### Million-line codebase refactor URL: https://agenticfeats.com/c/million-line-codebase-refactor-hermes-agent/ Hermes Agent split a million-line Python repo into 36 groups and dispatched 1,393 subagents in git worktrees to refactor it. Result: Non-test Python fell from 1,063,826 to 698,363 lines (34.4%), gateway/run.py from 34,847 to 5,512 lines, and functions over 300 lines from 192 to 2. - Agents: Hermes Agent - Evidence: grade A (Shown, artifact linked), Merged PR, benchmark gist, and before-and-after metrics table from the author - Outcome: Completed - Job type: Coding - Roles: Developer, Founder or owner - Industry: Software companies - Setup: Hermes Agent with a /goal objective and a hermes-agent-dev skill. Workers ran in git worktrees, checked against a frozen baseline, with tool schemas and CLI help output compared to the original. - Caveat: Reviewers caught removed public names and altered exception handling that tests missed. Fixes took two review rounds, plus more after merge. - Prompt (written by ${BRAND}, not the author's exact prompt): Set a standing goal: reduce non-test source by at least [percent]. Break up god files, unify duplicate helpers, replace long if/elif chains with dispatch tables, and keep public interfaces identical. Split the code into non-overlapping groups, assign each to a worker in its own git worktree, and require verified commits. - Evidence link: https://github.com/NousResearch/hermes-agent/pull/102117 - Source: https://nousresearch.com/refactoring-hermes-with-1393-agents, 2026-09-15 - Published: 2026-10-13 ### Shared team Hermes agent URL: https://agenticfeats.com/c/shared-team-hermes-agent/ Hermes Agent spread from the author's own use to their dev, co-founder, and a shared team agent named Robert. Result: Robert pulls context from calls, helps with approvals, and creates tasks in Multica, backed by shared memory. The non-technical co-founder reportedly gets more done. - Agents: Hermes Agent - Evidence: grade C (Claim, claim only), Author's own account; no numbers, repo, or demo - Outcome: Partial - Job type: Business - Roles: Founder or owner, Operations - Industry: Saas - Setup: Hermes Agent with shared memory built from an Obsidian knowledge base plus Hermes' own memory. Robert creates tasks in Multica and is reachable by the whole team. - Caveat: Still early. The author has not yet decided what a design team workflow should look like, and no numbers are given. - Source: https://x.com/niklas11891/status/2105541804267274335 by @niklas11891, 2026-10-01 - Published: 2026-10-13 ## Perplexity Comet ### Comet browser grocery and pizza orders URL: https://agenticfeats.com/c/comet-browser-grocery-and-pizza-orders-perplexity-comet/ Perplexity's Comet browser assistant built a charcuterie shopping list, added items to an Instacart cart, and tried to order a gluten-free pizza on Seamless. Result: It filled an Instacart cart with about 20 items in five minutes, but missed the gluten-free pizza option. - Agents: Perplexity Comet - Evidence: grade A (Shown, artifact linked), Reviewer's hands-on test with a stated item count and time; screenshots included - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: Comet AI browser with its built-in assistant sidecar. The author had to be logged into Instacart and Seamless. The assistant would not complete purchases. - Caveat: It defaulted to cheap or bestselling items, missed a gluten-free option, added unrequested extra cheese, and would not complete checkout. Summaries lacked source links. - Prompt (written by ${BRAND}, not the author's exact prompt): Create a shopping list for [dish or event], then add all the items to my [grocery service] cart at [store]. Do not check out. - Source: https://yahoo.com/life/articles/m-testing-perplexity-ai-browser-093500746.html, 2025-07-17 - Published: 2026-10-13 ### LinkedIn content creation automation URL: https://agenticfeats.com/c/linkedin-content-creation-automation-perplexity-comet/ Perplexity Comet's browser assistant created the author's LinkedIn content, replacing their manual content workflow. Result: The author says the automated LinkedIn content led to 5 new sales of their Gumroad product. - Agents: Perplexity Comet - Evidence: grade B (Numbers, a specific number stated), Author states 5 sales; the full article is paywalled - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Setup: Perplexity Comet browser and assistant, early access through a Perplexity Pro account, used to draft LinkedIn posts. - Caveat: The full story is paywalled and the sales are described as accidental, so the link to the posts is unverified. - Source: https://medium.com/@dheenmech007/here-is-how-perplexity-comet-browser-automated-my-linkedin-content-creation-bd72c0b353e9, 2025-08-03 - Published: 2026-10-13 ### Emailed AI news briefing URL: https://agenticfeats.com/c/emailed-ai-news-briefing-perplexity-comet/ Perplexity Comet searched major outlets for the day's AI news, picked the top five stories, and emailed the author a brief. Result: A five-story AI news brief arrived by email from one prompt; the author says plain LLMs only draft text and cannot send it. - Agents: Perplexity Comet - Evidence: grade A (Shown, artifact linked), Review with a demo video; the author reports two months of daily use - Outcome: Completed - Job type: Research - Roles: Personal - Setup: Perplexity Comet browser with its Gmail connector. The author notes the prompt can be saved as a shortcut or set as a recurring task. - Caveat: Comet can be slower than doing the task by hand, and long multi-step prompts often stop partway through. - Prompt (written by ${BRAND}, not the author's exact prompt): Search today's [topic] news across major outlets, pick the top 5 stories, and email me a short brief at [email address]. - Source: https://thecreatorsai.com/p/perplexity-comet-review, 2025-09-30 - Published: 2026-10-13 ### Booking a flight with Comet URL: https://agenticfeats.com/c/booking-a-flight-with-comet-perplexity-comet/ Perplexity Comet searched flights, then booked the one the author chose on Southwest.com while logged in, filling fare class and passenger details. Result: The agent completed the purchase after the author confirmed it, and Southwest sent a booking confirmation email moments later. - Agents: Perplexity Comet - Evidence: grade C (Claim, claim only), Journalist's first-person account; no screenshot or repo of the booking - Outcome: Completed - Job type: Personal and Family - Roles: Personal - Setup: Perplexity Comet browser, already logged into a Southwest.com Rapid Rewards account with a saved credit card. Flight search done first in Perplexity's conversational mode. - Caveat: The author calls it an entry-level experience; the agent did not yet know their preferences, and it paused for confirmation before purchase. - Prompt (written by ${BRAND}, not the author's exact prompt): Book the [flight option I picked] on [airline website] for [passenger name]. Pause before payment so I can confirm the fare and details. - Source: https://travelweekly.com/Travel-News/Travel-Technology/Agentic-AI-turns-searches-to-sales, 2025-11-05 - Published: 2026-10-13 ### Newsletter research with Comet URL: https://agenticfeats.com/c/newsletter-research-with-comet-perplexity-comet/ Perplexity Comet loaded NotebookLM with Reddit sources and analyzed Substack note performance for a newsletter author over a week of testing. Result: Comet added Reddit threads plus 10 Discover sources to NotebookLM in about 15 minutes unattended, then extracted 67 notes for analysis in 25 minutes. - Agents: Perplexity Comet - Evidence: grade B (Numbers, a specific number stated), Author's first-person write-up with screenshots; no repo or shared output. - Outcome: Partial - Job type: Marketing and Content - Roles: Marketing - Industry: Newsletters - Setup: Perplexity Comet browser agent working across tabs with NotebookLM's Add Source and Discover features and the Substack Activity page, driven by copy-paste prompts. - Caveat: Comet kept stopping after 15-20 notes and needed repeated prompting; the note extraction ended up slower than doing it manually. - Prompt (written by ${BRAND}, not the author's exact prompt): Research task: add [NUMBER]+ [SOURCE TYPE] about [TOPIC] to NotebookLM. Use Add Source to input URLs, then use Discover to find related sources on [TOPIC] and add them all. Goal: a notebook covering [WHAT YOU WANT TO ANALYZE]. - Source: https://cashandcache.substack.com/p/i-used-perplexity-comet-as-my-newsletter, 2025-10-28 - Published: 2026-10-13 ### API endpoint testing in Postman URL: https://agenticfeats.com/c/api-endpoint-testing-in-postman-perplexity-comet/ Perplexity Comet ran API endpoint tests in Postman after development, working from payload context the author supplied. Result: Comet generated multiple test payloads for different cases and produced a final report after the full test run. - Agents: Perplexity Comet - Evidence: grade A (Shown, artifact linked), Author's own account with a video attached; no counts given - Outcome: Completed - Job type: Coding - Roles: Developer - Setup: Perplexity Comet driving Postman, given enough context about the payload and other endpoint details. - Caveat: The author says it works when given enough context. No numbers or test counts are stated. - Prompt (written by ${BRAND}, not the author's exact prompt): Test the API endpoints for [project] in Postman. Here is the payload format and details: [payload and endpoint context]. Create payloads for different test cases, run them all, and write a final report of the results. - Source: https://x.com/ai_for_success/status/1995164144509784334 by @ai_for_success, 2025-11-30 - Published: 2026-10-13 ### Browser assistant for daily tasks URL: https://agenticfeats.com/c/browser-assistant-for-daily-tasks-perplexity-comet/ Perplexity Comet sent an email, set a calendar reminder, wrote a Google Drive post, and summarised a YouTube video. Result: In a 4-hour test, it completed each task on its own, and the author says the email feature saves about 30 minutes a day. - Agents: Perplexity Comet - Evidence: grade A (Shown, artifact linked), Author's own 4-hour test with a video walkthrough; the time saved is self-reported. - Outcome: Partial - Job type: Ops and IT - Roles: Founder or owner, Marketing - Setup: Install the Comet browser from perplexity.ai, import Chrome bookmarks, and connect Gmail so it can reach email, calendar and Google Drive. - Caveat: It sent the email unreviewed, used markdown instead of HTML, and began renaming Drive folders unasked, so the author stopped it. - Prompt (written by ${BRAND}, not the author's exact prompt): Write an email about [topic] and send it to [address]. Use HTML formatting. - Evidence link: https://www.youtube.com/watch?v=74o_4iQPDog - Source: https://juliangoldie.com/perplexity-comet-browser - Published: 2026-10-13 ## Claude computer use ### Gatsby lesson plan spreadsheet URL: https://agenticfeats.com/c/gatsby-lesson-plan-spreadsheet-claude-computer-use/ Claude with Computer Use downloaded The Great Gatsby, researched lesson plans and standards, and filled a spreadsheet on its own. Result: Ethan Mollick received a finished Gatsby lesson plan spreadsheet for high schoolers with no obvious errors, after delegating and walking away. - Agents: Claude computer use - Evidence: grade C (Claim, claim only), Secondhand account of Ethan Mollick's test, linked to his post - Outcome: Completed - Job type: Learning - Roles: Education - Industry: Schools - Setup: Claude with the Computer Use beta, accessed via API. It took screenshots, then drove the mouse and keyboard in a browser and spreadsheet app. - Caveat: The system is quite slow. Mollick found no obvious errors but says there may be some, and notes reliability concerns. - Prompt (written by ${BRAND}, not the author's exact prompt): Put together a lesson plan on [book] for [grade level] students. Break it into readable chunks, create assignments tied to [learning standard], and put everything in a single spreadsheet. - Evidence link: https://www.oneusefulthing.org/p/when-you-give-a-claude-a-mouse - Source: https://aiiq.substack.com/p/claude-can-now-use-a-computer, 2024-10-23 - Published: 2026-10-13 ### Restaurant search with computer use URL: https://agenticfeats.com/c/restaurant-search-with-computer-use-claude-computer-use/ Claude's computer-use agent searched Google for vegetarian restaurants near Bond Street and saved results to a notepad file on a remote machine. Result: It created the text file as asked, but after clicking a result it took a screenshot too early and hallucinated the answer. - Agents: Claude computer use - Evidence: grade A (Shown, artifact linked), Screenshots of the chat history and the saved text file - Outcome: Partial - Job type: Research - Roles: Personal - Setup: Anthropic's computer-use demo run in Docker with a Claude API key, started by one Terminal command and opened in a browser. - Caveat: The author calls it not currently useful based on this one run, because the agent acted before the page had loaded. - Prompt (written by ${BRAND}, not the author's exact prompt): Find 5 good vegetarian friendly restaurants near [location]. Save the results in a notepad file with the following information: name of restaurant, link to website, one line description. - Source: https://lovkush.substack.com/p/i-tried-anthropics-computer-use-agent, 2025-01-14 - Published: 2026-10-13 ### Flight booking with computer use URL: https://agenticfeats.com/c/flight-booking-with-computer-use-claude-computer-use/ Claude's computer use, running in a Docker container, opened Firefox and tried to book a New Delhi to Mumbai flight on MakeMyTrip. Result: Claude opened MakeMyTrip, cleared a sign-in popup after a second prompt, and began filling the search fields; no finished booking is shown. - Agents: Claude computer use - Evidence: grade A (Shown, artifact linked), Author's walkthrough with screenshots and a demo GIF; booking completion unclear - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: Claude 3.5 Sonnet computer use demo from Anthropic's quickstart, run in Docker with an Anthropic API key, controlling a virtual desktop in the browser. - Caveat: Beta tool. It struggled with the sign-in popup and hit errors until the author added a prompt to close it. - Prompt (written by ${BRAND}, not the author's exact prompt): Book my flights from [origin city] to [destination city] for [travel date]. - Source: https://rittikajindal.medium.com/my-first-hand-experience-with-claudes-computer-use-when-ai-takes-control-a0f78445a6c7, 2024-11-30 - Published: 2026-10-13 ### X feed research for posts URL: https://agenticfeats.com/c/x-feed-research-for-posts-claude-computer-use/ Claude computer use, sent from the phone via Dispatch, opened Chrome, read the author's logged-in X feed, and picked posts worth writing about. Result: In about 90 seconds it returned four posts with view and like counts, plus a suggested angle for each. - Agents: Claude computer use - Evidence: grade A (Shown, artifact linked), Author's screenshots of the Claude chat; blog post with no repo - Outcome: Completed - Job type: Research - Roles: Marketing - Setup: Claude Dispatch from the Claude mobile app, Claude in Chrome extension, Claude desktop app awake on macOS, author already logged into X in the browser. - Caveat: Research preview: Pro and Max only, macOS only, and complex tasks sometimes need a second try. The author still decides which posts become stories. - Prompt (written by ${BRAND}, not the author's exact prompt): Can you go to [site you're already logged into] and find [specific thing you'd normally spend 15 minutes looking for]? - Source: https://prosperinai.substack.com/p/claude-computer-use-workflow, 2026-03-26 - Published: 2026-10-13 ### Claude computer use on Mac URL: https://agenticfeats.com/c/claude-computer-use-on-mac/ Claude used its new computer use feature to take over a Mac, open the Chess app, and try to play a game. Result: Claude opened Chess and reported the board state, but never played a game, and half an hour of use drained the five-hour Claude Pro allotment. - Agents: Claude computer use - Evidence: grade C (Claim, claim only), Reviewer's first-hand account with screenshots; no repo or output file - Outcome: Partial - Job type: Setup and Meta - Roles: Personal - Setup: Claude consumer app on a Mac mini, Claude Pro plan, computer use toggled on in Settings, with accessibility and screen recording permissions and per-app access granted. - Caveat: Mac only, paid plans only, it needs screenshots of the whole screen, and it burns usage limits fast. Complex tasks failed. - Source: https://pcworld.com/article/3097542/claude-controlled-my-mac-for-half-an-hour-it-was-a-wild-worrisome-ride.html, 2026-03-24 - Published: 2026-10-13 ### Wallet data into spreadsheet URL: https://agenticfeats.com/c/wallet-data-into-spreadsheet-claude-computer-use/ Claude computer use pulled wallet data and dashboard numbers across the author's desktop, then summarized them in an open spreadsheet. Result: The task finished in 19 minutes, with numbers formatted in the spreadsheet and a screenshot trail of every step. - Agents: Claude computer use - Evidence: grade B (Numbers, a specific number stated), Author states 19 minutes and a screenshot trail; one photo, no repo - Outcome: Completed - Job type: Ops and IT - Setup: Claude with computer use turned on. No custom framework or loop. It worked across the author's real desktop, including a login page and a spreadsheet. - Caveat: Based on one messy prompt, self-reported with no artifact shown. The author says it handled wrong buttons and modals by itself. - Prompt (written by ${BRAND}, not the author's exact prompt): Pull [data type] from [dashboard or wallet tool], check the figures in [other tool], and put a cleaned-up summary into the open spreadsheet [sheet name]. - Source: https://x.com/rifat_ee/status/2068626729875181660 by @Rifat_EE, 2026-06-21 - Published: 2026-10-13 ## Operator ### Compiling a course list URL: https://agenticfeats.com/c/compiling-a-course-list-operator/ OpenAI's Operator browsed the author's online course site and compiled a list of courses with fees and descriptions while they got coffee. Result: Operator returned the requested list of courses, fees and descriptions without further input, saving the author a few minutes of manual admin work. - Agents: Operator - Evidence: grade C (Claim, claim only), Author's own account in a blog post; no output shown for this task - Outcome: Completed - Job type: Business - Roles: Education, Founder or owner - Industry: Course creators - Setup: OpenAI Operator, a research preview on the $200/month Pro tier, controlling a virtual browser through screenshots and text prompts. - Caveat: Low-stakes task. In the same post, Operator picked the wrong jeans size at checkout and could not reach paywalled course tests. - Prompt (written by ${BRAND}, not the author's exact prompt): Visit [website URL] and compile a list of all [courses/products] with their [fees/prices] and short descriptions. Return it as a clean list. - Source: https://leonfurze.com/2025/02/28/hands-on-with-openais-operator, 2025-02-28 - Published: 2026-10-13 ### Hands-on Operator test URL: https://agenticfeats.com/c/hands-on-operator-test/ Casey Newton tested OpenAI's Operator on walking tours, a Gatsby lesson plan, Instacart groceries, and an Uber price check. Result: Operator handled tour search and lesson plan but was slow; adding three grocery items took 15 minutes and an Uber price check took five. - Agents: Operator - Evidence: grade B (Numbers, a specific number stated), Reviewer's own timed tests; no repo or recording shown - Outcome: Partial - Job type: Research - Roles: Personal - Setup: OpenAI Operator, a research preview in the $200-a-month ChatGPT Pro tier, running in its own browser with screenshots and mouse actions. - Caveat: It uses its own browser, so it lacks logins and preferences, and some sites block it. Free ChatGPT did the simple tasks faster. - Source: https://platformer.news/openai-operator-ai-agent-hands-on, 2025-01-24 - Published: 2026-10-13 ### School financials from ProPublica URL: https://agenticfeats.com/c/school-financials-from-propublica-operator/ Operator searched ProPublica Nonprofit Explorer for independent schools and logged their financial data into a Google Sheet; o1 then built a dashboard. Result: Operator collected data for about 25 of 100 schools before the author stopped it; o1 then built a working chart dashboard from that data. - Agents: Operator - Evidence: grade B (Numbers, a specific number stated), Author's first-hand blog account with a demo video; no repo or sheet shared - Outcome: Partial - Job type: Research - Roles: Education, Operations - Industry: Schools - Setup: Operator browsing ProPublica Nonprofit Explorer, writing to a Google Sheet seeded with 100 school names from o3-mini-high. The sheet was pasted into o1, which wrote a Canvas chart. - Caveat: Operator was slow, rounded numbers inconsistently, and picked the wrong organization several times. The author stopped it at about 25 schools. - Prompt (written by ${BRAND}, not the author's exact prompt): Find the latest summary financials (revenue, expenses, assets) for each organization in my Google Sheet on [nonprofit database site]. Copy exact figures, unrounded, into the matching row. Check that the organization name matches before recording. - Source: https://fullstackeducator.substack.com/p/ai-agents-in-action, 2025-02-25 - Published: 2026-10-13 ### Operator concert and product research URL: https://agenticfeats.com/c/operator-concert-and-product-research/ Operator, OpenAI's browser agent, was tried on finding Berlin classical concerts and researching Anki competitors, using prompts and screen recordings. Result: Both tasks failed. The concert search returned three wrong concerts after 33 minutes, and the Anki competitor list had incorrect data. - Agents: Operator - Evidence: grade A (Shown, artifact linked), Author states times and failures, with screenshots and a screen recording - Outcome: Failed, with a lesson - Job type: Research - Roles: Personal - Setup: OpenAI Operator, a ChatGPT agent with its own VM and browser. The author wrote prompts, once drafted with GPT-4, and could not log in to accounts. - Caveat: The VM is not logged in to anything, and entering passwords was manual and clunky. The author calls it a research preview. - Source: https://antoine.fi/trying-openai-operator, 2025-01-29 - Published: 2026-10-13 ### Operator concert restaurant booking URL: https://agenticfeats.com/c/operator-concert-restaurant-booking/ Operator, tested by a member of Allie Miller's team, found a gluten-free restaurant near a Minneapolis concert venue and made a reservation. Result: Operator completed the restaurant search and booked the reservation, but a keto meal plan and a quilt pattern search came back weak. - Agents: Operator - Evidence: grade C (Claim, claim only), Author's team test report; one screenshot, no repo or public output - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: Operator from OpenAI/ChatGPT, run in the browser. Team members each had a $100 budget to test agents on real tasks. - Caveat: Fine for a decent option, weak for the best one. It read search terms too literally. - Prompt (written by ${BRAND}, not the author's exact prompt): Find a [dietary need] restaurant near [venue] in [city] for [date and time] and make a reservation for [party size]. - Source: https://aiwithallie.beehiiv.com/p/i-reviewed-10-ai-agents-and-what-actually-works, 2025-05-15 - Published: 2026-10-13 ### Operator review: tweets, news, builders URL: https://agenticfeats.com/c/operator-review-tweets-news-builders/ The author spent $200 on ChatGPT Operator and tested it on tweet analysis, AI news summaries, website builder pricing lists, and article summaries. Result: It returned duplicate and months-old tweets, summarized one news source, copied a single listicle, and stopped to ask permission mid-task. The author says productivity dropped. - Agents: Operator - Evidence: grade A (Shown, artifact linked), Author's own screenshots and a video of the runs; no repo - Outcome: Failed, with a lesson - Job type: Research - Roles: Design - Setup: ChatGPT Operator on the $200 plan, prompted directly with plain-language research and summarizing tasks. - Caveat: One person's week of tests on a version from March 2025; the source text is cut off before the screenshot and CAPTCHA tests. - Source: https://icons8.com/blog/articles/chatgpt-operator-review, 2025-03-27 - Published: 2026-10-13 ## Lindy ### Weekly newsletter drafting agent URL: https://agenticfeats.com/c/weekly-newsletter-drafting-agent-lindy/ Lindy AI agent researches weekly AI marketing news, drafts a newsletter plus social content, saves it to Airtable, and emails it every Monday. Result: The author built the agent in under 10 minutes; it delivers a finished newsletter draft by email each Monday at 9am. - Agents: Lindy - Evidence: grade A (Shown, artifact linked), Author's own screenshots and video; no sample newsletter output shown - Outcome: Completed - Job type: Marketing and Content - Roles: Founder or owner, Marketing - Industry: Newsletters - Setup: Lindy AI with a Monday 9am schedule trigger, a prompt, skills for Airtable, email, web search and AI writing, and the author's style guide as context. - Caveat: The author still uses Make.com for workflows needing precise control. The full template is offered only inside their paid community. - Prompt (written by ${BRAND}, not the author's exact prompt): Each Monday, research [topic] news from the past week. Write a newsletter draft for [audience] in the attached style guide. Also write LinkedIn posts, tweets and a video script. Save everything to Airtable and email me the newsletter. - Source: https://annikahelendi.substack.com/p/build-your-own-ai-newsletter-agent, 2025-06-08 - Published: 2026-10-13 ### Lindy for recurring ops work URL: https://agenticfeats.com/c/lindy-for-recurring-ops-work/ Lindy, connected to Slack, Notion, Gmail and Calendar, analyzed a guide library, built a staff meeting deck, and created a Notion guide idea from Slack. Result: Lindy produced a finished Monday staff meeting deck as a web link in 10 to 15 minutes and created a Notion page after approval. - Agents: Lindy - Evidence: grade A (Shown, artifact linked), Guide authors describe their own tests with screenshots; no repo or public output link - Outcome: Completed - Job type: Ops and IT - Roles: Founder or owner, Operations - Setup: Lindy connected to Slack, Notion, Gmail, Google Calendar and Drive. Used suggested prompts and prebuilt routines, with approval required for actions through connected accounts. - Caveat: Lindy works as a background teammate, so tasks take time. It asks for approval before creating items through a connected account. - Prompt (written by ${BRAND}, not the author's exact prompt): Review [set of documents or guides] from [time period] and summarize [trends, how coverage has changed, and what kind of content we make]. - Source: https://app.therundown.ai/guides/build-a-lindy-agent-that-never-drops-a-follow-up, 2026-09-04 - Published: 2026-10-13 ### Agent workforce for unstaffed functions URL: https://agenticfeats.com/c/agent-workforce-for-unstaffed-functions-lindy/ Ankor's CEO built Lindy agents covering lead research, technical support, content and marketing, and meeting prep, work too small to justify dedicated hires. Result: Ten agents were live in the first week and 20-30 run today. The CEO estimates at least a 5x return on investment. - Agents: Lindy - Evidence: grade B (Numbers, a specific number stated), Vendor case study quoting the CEO; screenshots are generic product UI, no independent artifact - Outcome: Completed - Job type: Business - Roles: Founder or owner, Operations - Industry: Software companies - Setup: Lindy agents connected to Gmail, Slack, HubSpot, Notion, Zoom and Google Sheets. The CEO built them himself, with no engineering team. - Caveat: The 5x return is the CEO's own estimate, published on the vendor's own case study page. - Source: https://lindy.ai/case-study/ankor - Published: 2026-10-13 ## Gemini agent ### Chrome auto browse five-task test URL: https://agenticfeats.com/c/chrome-auto-browse-five-task-test-gemini-agent/ Gemini's Chrome auto browse ran five real tasks for the author: hotel research, tax document downloads, grocery cart building, plumber quote forms, and subscription cancellation. Result: Hotel comparison, four tax PDFs, and an 18-item cart under $80 worked. Subscription cancellation failed. - Agents: Gemini agent - Evidence: grade B (Numbers, a specific number stated), Hands-on blog review with stated times and counts; screenshots are Google's - Outcome: Partial - Job type: Personal and Family - Roles: Personal - Setup: Chrome auto browse (Gemini 3) in preview on a Google AI Pro or Ultra plan, working inside the author's already logged-in browser sessions. - Caveat: It cancelled the wrong subscription tier and stalled on two cancellation modals. Two cart items were the wrong brand. Preview is US-only and paid-plan only. - Prompt (written by ${BRAND}, not the author's exact prompt): Get me written quotes from three local [service providers] for [job]. Use the contact forms on their sites. Do not send anything until I review. - Source: https://blog.imseankim.com/chrome-auto-browse-gemini-agentic-hotel-form-fill-google-io-2026, 2026-05-19 - Published: 2026-10-13