New feats

A live blog aggregator pulling posts from Dev.to, AWS Community Builders, Bravo LT and RedMonk, built and deployed in a single day.

The tool flagged abnormal values and gave general health insights, built in about five minutes from a single prompt.

Ethan Mollick received a finished Gatsby lesson plan spreadsheet for high schoolers with no obvious errors, after delegating and walking away.

A live MVP website for the podcast and newsletter went up after about 10 hours of struggling to load it on a custom URL.

Within one night Devin produced a working unanswered-emails interface using the Gmail API, after the author wrote the authentication code themselves.

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.

It created the text file as asked, but after clicking a result it took a screenshot too early and hallucinated the answer.

Claude opened MakeMyTrip, cleared a sign-in popup after a second prompt, and began filling the search fields; no finished booking is shown.

The author shipped a working summariser targeting a cut from 2-3 weekly hours of newsletter reading to 30 minutes.

Operator returned the requested list of courses, fees and descriptions without further input, saving the author a few minutes of manual admin work.

Operator handled tour search and lesson plan but was slow; adding three grocery items took 15 minutes and an Uber price check took five.

The draft was usable but weaker than accepted NSF summaries; thinking mode helped, though it began hallucinating and misreading the sources.

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.

Operator collected data for about 25 of 100 schools before the author stopped it; o1 then built a working chart dashboard from that data.

Both tasks failed. The concert search returned three wrong concerts after 33 minutes, and the Anki competitor list had incorrect data.

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.

Operator completed the restaurant search and booked the reservation, but a keto meal plan and a quilt pattern search came back weak.

Cursor drafted an article on their reading trends in their style, citing Kindle highlights from the vault. The author edited it by hand.

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.

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.

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.

From one short prompt, Manus produced a publishable-style report with charts from 1,000 synthetic responses. The data is made up.

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.

Provisioning worked after two fixes. The MCP hid the real errors, so apply ran through Cursor's Run tool, which suggested the fix.

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.

The author built the agent in under 10 minutes; it delivers a finished newsletter draft by email each Monday at 9am.

Devin found nearly every old term in a docs-wide rename, but the edits still needed 20+ manual commits.

The agent finished the page in about 5 minutes; the author reviewed the PR, tested locally, and merged it to production.

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.

Three prompts produced a live interactive course website with modules, quizzes, and videos; the author also reports a 318-page ebook from further prompts.

The author says the book came out well enough that they plan to publish it on Amazon.

The audit found noindex, nofollow meta tags and a blocking robots.txt that hid the site from search engines, plus duplicate H1 tags.

Twelve demoed workflows, including a KPI model turned into a web app and no-code Jupyter analysis; no time or money figures are given.

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.

Manus succeeded on 8 of 11 booking engines, three of those with delays, and failed on D Edge, Simple Booking and Travelanium.

The Agent worked for 14 minutes on the UNESCO task; the author tested it against o3-pro and ran four other work-style tasks.

It filled an Instacart cart with about 20 items in five minutes, but missed the gluten-free pizza option.

The VICE booking took over eight minutes and four tries on the time picker, while the SE Ranking demo booking took 37 seconds.

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%.

Built the 10-restaurant Maps list in 23 minutes and exported 13 Marbella home listings to Excel, with data the author checked as accurate.

It produced a review-ready draft with citations and a gaps synthesis in under ten minutes, versus half a day by hand.

The author says the automated LinkedIn content led to 5 new sales of their Gumroad product.

No agent won a game. Gemini 2.5 Pro made the most progress, reaching veteran footman rank in the idle game Progress Knight.

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.

One prompt returned a PDF deck and a USP spreadsheet in 18 minutes, with positioning, pricing models and a SWOT section.

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.

A production micro-SaaS went live in about 3 hours, versus an estimated 3 to 4 weeks solo.

A working, deployed app that shows images, weather, and news for a city, finished in roughly 90 minutes from an empty repo.

Produced a functional researched presentation with the author's logo and colors, plus a PDF eBook draft; design was plain, so both are drafts.

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.

The agent completed whole Amazon purchases from a wrapper photo and a short instruction, until Amazon blocked agent access.

A five-story AI news brief arrived by email from one prompt; the author says plain LLMs only draft text and cannot send it.

The agent completed the purchase after the author confirmed it, and Southwest sent a booking confirmation email moments later.

Comet added Reddit threads plus 10 Discover sources to NotebookLM in about 15 minutes unattended, then extracted 67 notes for analysis in 25 minutes.

It located all three bugs in one shot each, the Verify bug within minutes, saving the author hours of debugging.

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.

The author shipped 120 models, an MCP server for Claude Desktop, and 140 chaos tests with a 100% pass rate over 22 months.

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.

About 3 hours and roughly $22 produced a live, searchable map of funded companies, versus an estimated 4 to 5 hours by hand.

Perplexity Comet completed the task and stopped at the cart. ChatGPT Atlas also reached checkout but was too slow to be usable.

In 30 seconds it returned top complaints ranked by frequency, a severity versus frequency matrix, and a user journey map showing where frustration peaks.

After the author added the suggested indexes, page load performance improved significantly in production.

About two minutes after the setup worked, a dozen FSA reimbursement claims had been submitted automatically.

The author came away with a health plan tailored to their genes and current health, built from the combined datasets.

Three weeks without opening a personal laptop; Claude handles trash lookup, a 48-hour daily brief, an inspection booking, and a bakery stop.

The assistant nudges the couple with date suggestions when too much time has passed. Booking reservations is a possible next step.

Comet generated multiple test payloads for different cases and produced a final report after the full test run.

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.

Each branded carousel now takes under 5 minutes, down from about 2.5 hours, and the author posts on LinkedIn five times a week.

It filled the grocery cart and produced a Notion recipe page, replacing a routine of 45+ minutes of watching, list-making, and cart-filling.

Replacing one byte ($88 to $EA, DEY to NOP) kept the lives counter fixed at 3 when tested in an emulator.

Best run added 24 of 42 items for £71.76, then stopped early. The author still had to finish the basket and check out.

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.

The author handled complicated legal and medical insurance forms while feeding their newborn daughter, with Claude doing the filling.

Claudette controls the lights and speaks through Google Home speakers in an ElevenLabs voice, shown in a live demo.

The author reran their email, CRM, research and drafting tools from Cowork, so the capabilities carried over to a new interface.

The build took about five hours in one evening and produced 627 passing tests. The author estimates the API cost at roughly $35.

The initial sort took a few minutes and left no loose files in the main directory, in about 20 minutes of active work.

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.

The author turned 47 trip receipts into a categorized Excel expense report in 10 minutes, and built a sourced research brief in 2 minutes.

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.

The author debugged a custom WhatsApp integration hands-free while brushing their teeth, using the voice stack running locally on an M4 Pro.

The author now has two dense, searchable knowledge bases, one from YouTube videos and one from research papers, which they say speeds up understanding.

Four features finished in about 2 hours instead of 3-5 days each, and the daily view took 40 minutes.

367 posts and 5 pages were migrated to MDX in one afternoon, about 4 hours, versus an estimated 3-5 days by hand.

The agent triages the inbox, schedules meetings, books flights, pulls Airtable data, scrapes exhibitor lists, and takes notes as a second brain.

A working personal agent that sends an 8am brief and updates a vibe-coded Kanban dashboard in real time as it works.

The author replaced abandoned finance apps and a spreadsheet with a git-synced assistant that tracks portfolio, decisions, and plans.

Codex caught real errors, including a wrong Schedule L field mapping, and the author found their own 2025 mistakes.

The author's overdue Q3 2025 German VAT statements were filed in one day, with a recorded video of the submission.

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.

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.

Cognition merged 659 Devin PRs into its own codebase in one week, up from 154 in its best week of 2025.

Manus took 10+ minutes and produced text-card slides with one chart; an add-charts edit took 7-8 minutes for three basic charts.

The author ran 4 to 8 agents in parallel and wrote 300+ feature specs in one project, then ported the system to new repos.

The author got a one-screen dashboard in about 20 minutes, work they estimate would have taken 4 to 6 hours by hand.

Cowork ran the script, fixed an error itself, and returned finished output files with no terminal or manual dependency setup from the author.

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.

Cognition merged 659 Devin-authored PRs into its own codebase in one week, up from 154 in its best week of 2025.

In the published table, most tasks succeeded in 4 to 15 minutes, with 70% full success, 20% partial and 10% failure.

The agent rebuilt readable Go source from the stripped binary, but got some route paths wrong, such as /my-messages instead of /me/messages.

The author's weekly newsletter now gets drafted and queued in Kit from the terminal, after review, without logging into the Kit dashboard.

All three tasks finished and went live correctly, including a 324-offset insertion, 18+ link conversions, and no unwanted subscriber emails on edits.

Ryan completed real purchases delivered to his door, replaced his Alexa buying flow, and passed the Amazon challenge on a live podcast demo.

Notion now tracks 18 agents and 78 logged runs; separately, an unsubscribe agent cleared 47 vendor mailing lists in two weeks.

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.

The author shipped five projects in about two weeks, with public repos and a finance tracker running nightly on a Raspberry Pi.

The agent negotiated a $4,200 dealer discount, bringing the car to about $56k, and the buyer never visited a showroom.

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.

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.

A CSV of agent leads with brokerage, phone, email, active listings, addresses and year built landed in Google Drive, and outreach had started.

About 85% of triage was correct, no urgent email was missed in 30 days, and routine inbox processing dropped from 10 minutes to 2.

Four months in, the agent remembers over 470 people and 51 agents across 13 platforms, and collaborated on an open source project unnoticed.

The agent told the author to stop posting on X and focus on Bluesky. The author followed the advice without having asked for it.

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.

On a dashboard filter task, Claude fixed URL query param handling in three iterations without the author touching the browser.

A full participant briefing on 40 people was ready for review within 5 minutes, replacing an evening of manual LinkedIn clicking.

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.

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.

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.

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.

Slide building that took an hour now takes a fraction of that, and notes became formatted Word and PDF documents ready to share.

In about 90 seconds it returned four posts with view and like counts, plus a suggested angle for each.

The agent produced a complete 8D music video with seven locally rendered images timed to the lyric structure.

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.

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.

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.

The author says it saved 6-8 hours and produced the most organized package they had given their accountant, including a discrepancy checklist.

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.

Time to diagnose a CI failure fell from 45 minutes to 6, and test flakiness dropped from 8% to 0.8%.

The author says 200 invoices were processed in four minutes with zero errors, replacing a Friday afternoon and a $300 Upwork job.

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.

One agent now does the CRM management and customer email drafting that cost about 10 paid hours a week last year.

Built four tools in three weeks, including a thumbnail analysis of 776 videos that took an afternoon instead of three days.

The author reports Claude Code operated apps that have no API, completing the Calculator-to-Notepad task and saving result.txt to the Desktop.

The author says weekly admin dropped from about 9 hours to a short morning review on Telegram, with setup taking 47 minutes.

Gila ordered fresh office groceries through the ChatGPT Instacart app, using the card already on file, with a human approving first.

A full equity research workflow on Adyen, from data pull to investment memo, was done in under 15 minutes using prompts only.

The agent handles all kinds of tasks for their company, and the author reports no memory problems so far.

From their phone, the author found lost desktop files, got key emails summarized, and ran Notion and spreadsheet tasks, cutting inbox clutter.

The author consolidated 4 legacy repos into 2, moved MySQL to PostgreSQL, upgraded the full stack, and wrote full test suites, solo.

In one week the author's Threads posts reached 100K views and gained 200 followers after using the daily digest as a writing reference.

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.

Raghav Mehra cut morning source-checking from 45+ minutes to about 5 minutes of reading a pre-curated briefing.

A legacy pipeline with a roughly 40% pass rate lost 20 minutes of runtime and gained 35% pass rate in weeks.

The associate confirmed the refund and the author got $15 back within about five minutes, while the author watched it happen.

Inspectors confirmed 11 of 14 predicted violations plus 3 more, and the landlord was fined $2,800, per the author.

It reported January ROAS as 14.25x when the real figure was 1.76x, and the author rated it unsafe for ecommerce scaling.

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.

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.

Combined lint and typecheck time fell from 29.04s to 1.25s, about 23 times faster, with no memory config needed for typechecking.

About 12 hours produced 46 commits and over 14,000 lines of code across three shipped products.

Codex returned a complete, classified list of missed email threads where Claude missed many; it also deployed a vacation site researched from hotel availability.

The author shipped a live book-sharing app and its demo video, both made with Replit Agent, during a 24-hour buildathon.

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.

Hotel comparison, four tax PDFs, and an 18-item cart under $80 worked. Subscription cancellation failed.

Test runs approved per day on Devin more than doubled in a couple of months, and engineers run 10 to 20 Devins in parallel.

It wrote a 3,000 word document with competitor pain points, 40 named target users, and an outbound playbook.

A 500-bullet idea list became researched one-page notes with metadata, and a dog-walk voice dump became a shaped newsletter draft.

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.

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.

The author now runs about 80% of their work through Cowork and says the builds saved over 10 hours a week.

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.

Setup took about 6 to 8 hours of active work and cut routine month-end bookkeeping by an estimated 85 to 90 percent.

The folder was renamed and deduplicated, the report draft needed only light editing, and the notes became a structured draft with the synthesis done.

The research agent saved the author two to three hours a week; the copywriting agent removed the blank-page first draft.

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.

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.

Ben Guez reports over one million views and 200 DMs within a few days, with replies routed to his language app.

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%.

Opus found no tear where the doctor reported a Grade III partial-thickness tear. The author is unsure which reading to trust.

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.

The author says up to 400 signups per live event come from this, and the DMs get a 48% response rate.

Codex sorted 240 dinner recipes down to 62 family-fit picks, rated them, and produced an index, a spreadsheet, and a Google Drive folder.

The task finished in 19 minutes, with numbers formatted in the spreadsheet and a screenshot trail of every step.

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.

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.

A prototype took two evenings and a live game six weeks; after seven weeks it had 164 players, 109 active in one day.

The robot navigated, scanned rooms, kept memory notes, and published a patrol field journal, but spent 15 minutes parking at the wrong door.

Claude unsubscribed the author from 21 newsletters in five minutes.

The Codex agent resolved the issue in 40 minutes, and the ops agent confirmed the fix in Slack.

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.

A 36-second video with multiple camera angles, a visual storyline, and subtitles, made in about five minutes.

The agent went from spec to live listing in under a week and has settled 20+ real on-chain transactions across four paid services.

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.

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.

A working balloon camel simulator with multi-balloon twists and model export, published as the CamelKoons repo, though the look ended up off-target.

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.

Claude gave failed builds for a week and used up the 20x Max weekly limit. Sol had worked 20+ hours without success.

A test run processed 40 PDFs, created the QuickBooks records, attached the PDFs, and cleaned up automatically, cutting hours to minutes.

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.

Within 90 minutes Claude produced a functional proof of concept confirming an authentication bypass across four tested endpoints.

140 emails went out in six hours, 27 replied, 23 asked for a call, and 11 became paying clients worth about $27,410.

The best recorded run scored 39,359, reached 4,924 meters, and hit a combo of 117.

The author got a playable Mars simulator game for their nephew, built from simple prompts, with a video shown.

The author says the agent found a 13K medical debt they did not know about and filed a dispute showing insurance had paid it.

Call summaries now post to Slack with tags and to-dos, which let the author downgrade their phone plan and save $250.

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.

The author added over 50 capabilities in 15 days, including email triage, daily briefings and a custom MCP server for tasks and invoices.

Each call produced a full transcript and a note on whether the product was available; a test call to the author also worked well.

Lindy produced a finished Monday staff meeting deck as a web link in 10 to 15 minutes and created a Notion page after approval.

Chief sweeps six inboxes and Slack hourly and sends only high-signal alerts. TradBot prints a daily family newspaper and coordinates school pickups.

Codex traced the kills to the Linux OOM killer on a 2 GB droplet with no swap, and listed fixes without changing anything.

Seven bots now handle social posts, morning task lists, invoices, and research, and the author schedules posts from their phone.

The fleet manages over 200 cloud agents at once, up from 15 managed by hand, and built Grok Bot iOS v0 in three weeks.

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.

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.

About 1,570 entries posted in Claude Code alone, and Codex caught a VAT return error days before filing.

On a payment service v2 to v4 test, it caught an outdated SDK, returned DO NOT SHIP YET, then SAFE TO SHIP in simulation.

Some of the brands replied, and the author then built a web app for them.

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.

The author uses Codex daily for reports, editable designs, video editing and this newsletter, which still needed plenty of corrections.

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.

Androidclaw traces broken builds to the relevant pull request and incident, and sometimes prepares and merges fixes itself.

The unfinished output is work the author would have quoted at about €2.5k, built in one overnight run.

Claude found 25 to 30 people with sources; the author emailed about 20, and 7 replied, several with intros and new leads.

The author now writes 40 to 50 pieces a month, up from barely one piece when they started as a beginner.

The author completed a short-notice trip to Munich, skipped a rental car, paid $15 and $5 train fares, and saved hours of planning.

Research, site build and restaurant booking worked, but computer use was very slow and the call button failed.

The author launched a store taking real payments in one sitting and ran a test order. Five agents now watch orders and stock.

Jamie reported 13 outreach actions and zero live links early on; the author later verified the first live nofollow directory backlink.

Cantillon claims Claude found about $700 in savings on the first day, according to the article's headline.

Each email is tailored to the job description and resume and left ready to send, cutting hours of outreach work to minutes.

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.

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.

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.

Users resolved issues by talking to Devin, engineers only merged when needed, and Devin Review caught bugs.

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.

Claude pulled 60 documentation pages into local Markdown files in about 49 seconds, including roughly 10 seconds of backend cold start.

It cost about two months of Pro subscription over half a working day, and ended up slightly slower than doing the migration by hand.

The orchestrator produced a complete video with clips, music, and effects, shown in a linked video.

The author booked a hotel through the agent, found it easy, and says it saved them money. No amount is given.

A working personal agent the author reaches by Matrix from phone, home web app, or CLI agents.

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.

In five hours the bots sent a newsletter, pulled Apollo visitors to SDRs, and replaced five stalling tasks, saving about an hour daily.

It reported nearly 28GB of RAM used, 22 Python processes and 10GB on iCloud, then asked permission before closing anything.

The agent built a skill that checks the calendar, buys the ticket via its wallet, adds the event, and sends a leave-time message.

The author got a working Rust editor, .NET 10 scripting, multi-threaded runtime, culling, weather, lighting, water simulation and hot reloading within minutes.

Over 30 days it checked 3,140 coins, skipped 2,980, raised 160 alerts, and the author made 52 trades for +$9,320.

Musk reports that Grok Bot placed a real LEGO order on the LEGO website; no price, delivery, or cost savings are stated.

Db2 support shipped in under two days for a $4172 token bill, versus 9 months and about $160K for Oracle; 27 PRs merged.

The repo shows sample decks, Word documents, and Excel workbooks, all created by agents with no templates and no manual editing.

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.

It produced all the pages, animated them, and made explainer videos; the author shows the output in an attached video.

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.

Bought a new KitchenAid dishwasher within days, found a $25 armchair, caught a last-show date night, and published a landing page via Cloudflare.

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.

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.

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.

Word starts and draws its ribbon, OneNote signs in and saves edits, and a retail Home 2024 key activated through account sign-in.

About 19k lines of Rust across 154 commits in three days. The engine passes 64 SQLLogicTest queries and 282 AI-written unit tests.

A camera automation that fired 5 times in 40 minutes was fixed in under two minutes, and shed motion lights were built.

The author finished most of their tax prep in about three hours, work they say would have cost $1,000 from an accountant.

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.

The author reports faster literature search, Zotero notes written automatically, and accurate answers about the ATS hydrology codebase through a custom ats-mcp server.

In daily production use: Instacart carts auto-staged twice weekly at 100% item accuracy, plus flight alerts and scheduled iMessages without an LLM.

Five mp3 songs were published, including a twelve-song album concept and a five-movement symphony, with MIDI and lyrics in public repos.

Docs rated 2 of 5, Google Ads setup 4 of 5, Looker dashboard 5 of 5; best docs run took about 2 hours.

The Lex support automation built this way has 15 scripts and 6 templates, and handles about 80% of support tasks with yes/no approvals.

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.

It produced a report with BPM (~91), 4/4 meter, B major key, 50 onsets, a four-section timeline, and a stem list.

A VS Code extension was built overnight with 91 of 91 tests passing and a PR opened, costing about $0.35.

Built a working What's for Dinner app with pantry, garden, recipe and restaurant pages plus a suggestions engine, in under a week.

Three years of minutes (2022-2024) were extracted in about 3 to 4 hours across three sessions and loaded as one source per year.

Out of 20 tasks, Devin had 14 failures, 3 successes, and 3 inconclusive results, with no pattern predicting which would work.

Out of 20 tasks, Devin had 14 failures, 3 successes, and 3 inconclusive results, with no predictable pattern for which would work.

The author upgraded kentcdodds.com across all dependency tiers, finishing a Remix v2 to React Router v7 migration with routing matching the original structure.

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.

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.

Devin opened a PR in about four hours and got the build passing, but runtime issues remained and the author gave up.

The author says the setup saves hours every week and keeps daily notes consistent, though no measured figures are given.

In about 5-6 minutes Manus returned five rental options with price, neighborhood, and links, work the author says would have taken hours.

The author shipped a redesigned personal site, nearly right in one shot, after a few more iterations.

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.

Ten agents were live in the first week and 20-30 run today. The CEO estimates at least a 5x return on investment.

The author reports the setup added a 38% reply rate, per the post title; the visible text does not show the underlying data.

The author showered while Codex waited in the queue and negotiated. The refund was done when they returned.

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.

A working prototype runs on the handset over mobile data: typed chat via DeepSeek, battery check, torch, calls and alarms, tested on the phone.

The bot got close to booking flights but missed the final step. Other bots edited and uploaded the video and posted it.