Claude Code use cases: 74 real examples
Agentic Feats lists 74 sourced examples of Claude Code in real work; 45 link the artifact; most common job type: coding (26). Newest first.
- Feats
- 74
- Shown
- 4561%
- Industries
- 8
- Updated
- Oct 14
Claude Code is Anthropic's coding agent. It works in a terminal or an editor, reads and changes the files of a project, and runs commands, so people also point it at work that is not code: books, taxes, music, and research.
56 of 74 runs completed, 16 only partly worked, and 2 failed with a lesson. The partial and failed runs stay listed, each with what went wrong.
Claude Code docs ↗EvidenceOutcomesBy job typeQuestionsWhat AI agents got wrong
How strong is the evidence?
counted from 74 feats
| Grade | What it means | Feats | Share |
|---|---|---|---|
| Shown | artifact linked | 45 | 61% |
| Numbers | a specific number stated | 14 | 19% |
| Claim | claim only | 15 | 20% |
How we grade. A grade says how much proof the source gives, not how good the result was.
What worked, what partly worked, what failed
counted from 74 feats
| Outcome | Feats | Share |
|---|---|---|
| Completed | 56 | 76% |
| Partial | 16 | 22% |
| Failed, with a lesson | 2 | 3% |
Runs that fell short
ResultNo agent won a game. Gemini 2.5 Pro made the most progress, reaching veteran footman rank in the idle game Progress Knight.
Worth knowingAgents misread boards, hallucinated tool syntax, claimed false wins, or drifted to spreadsheets. Opus 4.1 reported a Mahjong win without matching a single pair.
ResultClaude gave failed builds for a week and used up the 20x Max weekly limit. Sol had worked 20+ hours without success.
Worth knowingNeither agent delivered a working viewer. The author deleted the Claude project and says swearing at the agent speeds up compliance.
ResultWithin 90 minutes Claude produced a functional proof of concept confirming an authentication bypass across four tested endpoints.
Worth knowingA safety guardrail briefly blocked log analysis and swapped the model; the author notes the full exploit chain claim was not entirely accurate.
ResultThe author shipped five projects in about two weeks, with public repos and a finance tracker running nightly on a Raspberry Pi.
Worth knowingThe author rarely uses the CRM, and the local 3B model handles note queries only decently and may break as notes grow.
Every partial and failed run, across all agents, is on What AI agents got wrong.
By job type
The best-scored completed feats with the artifact linked, in the largest job types.
Coding
26 feats in codingResultClaude pulled 60 documentation pages into local Markdown files in about 49 seconds, including roughly 10 seconds of backend cold start.
ResultAcross 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.
ResultIt found multiple remotely exploitable kernel bugs, including an NFS heap overflow undiscovered for 23 years. Five fixes or reports are linked in the post.
Finance and Admin
11 feats in finance and adminResultA test run processed 40 PDFs, created the QuickBooks records, attached the PDFs, and cleaned up automatically, cutting hours to minutes.
ResultThe 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.
ResultThe 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.
Creative
7 feats in creativeResultFive mp3 songs were published, including a twelve-song album concept and a five-movement symphony, with MIDI and lyrics in public repos.
ResultClaude Code produced a playable wave-based game with four enemy types; the creator reached wave 10 on camera and said it matched their vision.
ResultIt produced a report with BPM (~91), 4/4 meter, B major key, 50 onsets, a four-section timeline, and a stem list.
Questions about Claude Code
- What can Claude Code do?
- The 74 feats here cover 11 job types. The largest are coding (26), finance and admin (11), and creative (7). Coding is its home ground, but people also use it for admin, creative, and research work.CodingFinance and AdminCreative
- Can non-developers use Claude Code?
- Yes, though it is set up and run from a terminal, which takes some getting used to. 5 feats here are personal and family projects, and several come from people who say they do not code.Personal and Family
- What is Claude Code bad at?
- 16 of the 74 feats only partly worked and 2 failed. Their caveats point to weak spatial reasoning, judgment calls that still need a person, long runs that drift, and runs that reported work as done when it was not.What AI agents got wrong
- Is Claude Code good for bookkeeping and finance?
- 11 feats here are finance and admin work, such as reconciliations, invoices, and tax prep. Where the runs fell short, the misses were judgment calls, like personal versus business spending or local tax rules, so a person should still sign off.Finance and Admin
- How reliable are these examples?
- Each feat carries an evidence grade. Shown means the artifact itself is linked, Numbers means the source states a specific number, and Claim means a claim only. Of the 74 feats for Claude Code, 45 are Shown, 14 Numbers, and 15 Claim.How we grade
- How does Claude Code compare to Codex or Cursor?
- We do not rank agents. For scale only, Codex has 23 feats and Cursor 19 feats, graded the same way, each on its own page.CodexCursor
- How are feats chosen?
- A pipeline reads public posts, repos, videos, and articles every day, keeps the ones where someone says what an agent actually did, and writes each card in our own words. Every card is checked automatically, and a person reviews the flagged ones. No vendor pays for a place.MethodologyAbout
- How can I submit a Claude Code result?
- Send the link to the post, repo, video, or article on the submit form, with the agent and the result in one line. A person reviews every submission before it is published.Submit a Claude Code feat
Every count here is read from the feats at the time the page is built.
A live blog aggregator pulling posts from Dev.to, AWS Community Builders, Bravo LT and RedMonk, built and deployed in a single day.
Blog aggregator in one day
Claude Code built and deployed a Vue 3 blog aggregation site from a year-old requirements document, rewritten as a focused prompt.
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.
Canvas LMS tool via MCP
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.
The audit found noindex, nofollow meta tags and a blocking robots.txt that hid the site from search engines, plus duplicate H1 tags.
SEO audit of a blog
Claude Code, using the Puppeteer MCP server, audited the author's Nuxt blog starter and wrote its findings into a markdown report.
Twelve demoed workflows, including a KPI model turned into a web app and no-code Jupyter analysis; no time or money figures are given.
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.
No agent won a game. Gemini 2.5 Pro made the most progress, reaching veteran footman rank in the idle game Progress Knight.
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.
It located all three bugs in one shot each, the Verify bug within minutes, saving the author hours of debugging.
Debugging a cryptography implementation
Claude Code debugged a fresh Go implementation of the ML-DSA signature algorithm, finding why Verify rejected valid signatures and two separate Sign bugs.
The author shipped 120 models, an MCP server for Claude Desktop, and 140 chaos tests with a 100% pass rate over 22 months.
Mental models framework built with agents
A solo founder gave Claude, ChatGPT, Windsurf, and Cursor defined roles to build and deploy a 120-model mental models framework.
About two minutes after the setup worked, a dozen FSA reimbursement claims had been submitted automatically.
FSA claims from Amazon invoices
Claude Code with Playwright MCP read Amazon PDF invoices and filed healthcare FSA reimbursement claims on the benefits portal.
The author came away with a health plan tailored to their genes and current health, built from the combined datasets.
Genetic and bloodwork health plan
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.
Three weeks without opening a personal laptop; Claude handles trash lookup, a 48-hour daily brief, an inspection booking, and a bakery stop.
Personal assistant via Claude Skills
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.
The assistant nudges the couple with date suggestions when too much time has passed. Booking reservations is a possible next step.
Date night planning assistant
Claude Code, set up as an executive assistant, suggests date nights and double dates using the author's calendar and place lists.
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.
Self-hosted home server setup
Claude Code, installed on a Beelink mini PC over SSH, set up Docker, Caddy, and a set of self-hosted services at home.
Get five like these every week
The week's strongest feats, each with its receipt, plus a short roundup of AI news. Free.
You're on the list.
A free weekly newsletter. See what's in it
Each branded carousel now takes under 5 minutes, down from about 2.5 hours, and the author posts on LinkedIn five times a week.
LinkedIn carousels via Claude Skills
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.
Replacing one byte ($88 to $EA, DEY to NOP) kept the lives counter fixed at 3 when tested in an emulator.
Unlimited lives in River Raid
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.
The author reran their email, CRM, research and drafting tools from Cowork, so the capabilities carried over to a new interface.
Claude Code tools moved to Cowork
Claude Cowork read the author's folder of personal scripts built in Claude Code and loaded them into its memory in eleven steps.
The build took about five hours in one evening and produced 627 passing tests. The author estimates the API cost at roughly $35.
Data analysis agent suite
Claude Code built a six-agent data analysis suite from a 713-line spec the author drafted with another AI.
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.
Parallel agents build a C compiler
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.
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.
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.
The author debugged a custom WhatsApp integration hands-free while brushing their teeth, using the voice stack running locally on an M4 Pro.
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.
367 posts and 5 pages were migrated to MDX in one afternoon, about 4 hours, versus an estimated 3-5 days by hand.
WordPress blog to MDX
Claude Code, run as a terminal agent, pulled every post from the author's WordPress REST API and converted them to local MDX files.
The agent triages the inbox, schedules meetings, books flights, pulls Airtable data, scrapes exhibitor lists, and takes notes as a second brain.
Personal assistant agent for daily work
The authors built a personal agent with Claude Cowork or Claude Code, connected through Rube to their email, calendar, Slack, Airtable and other apps.
The author replaced abandoned finance apps and a spreadsheet with a git-synced assistant that tracks portfolio, decisions, and plans.
Personal finance assistant in Markdown
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.
Codex caught real errors, including a wrong Schedule L field mapping, and the author found their own 2025 mistakes.
S-Corp taxes with two agents
Claude Code wrote Python scripts that compute the author's S-Corp tax forms from QuickBooks and bank data, while Codex reviewed the code.
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.
Portfolio optimization plan from brokerage CSVs
Claude Code parsed the author's brokerage CSVs and notes, then built a phased portfolio optimization plan with tax analysis and fund swaps.
The author ran 4 to 8 agents in parallel and wrote 300+ feature specs in one project, then ported the system to new repos.
Parallel coding agents with specs
Claude Code agents run in separate tmux windows as planner, worker and PM, driven by numbered Markdown feature specs and six slash commands.
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.
YouTube research into Obsidian graph
Claude Code searched YouTube, loaded 20 videos into NotebookLM, asked cited questions, and saved everything to Obsidian as linked notes.
The author's weekly newsletter now gets drafted and queued in Kit from the terminal, after review, without logging into the Kit dashboard.
Newsletter drafts via Kit API
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.
The author shipped five projects in about two weeks, with public repos and a finance tracker running nightly on a Raspberry Pi.
Personal apps on a Raspberry Pi
Claude Code built five personal tools for a non-coder, including a multiplayer game, a WhatsApp scheduler, and a finance tracker.
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.
Tax organizer and return review
Claude in Chrome and Claude Code filled a web-only tax organizer, then Claude Code reviewed last year's return for errors.
Four months in, the agent remembers over 470 people and 51 agents across 13 platforms, and collaborated on an open source project unnoticed.
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.
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.
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.
On a dashboard filter task, Claude fixed URL query param handling in three iterations without the author touching the browser.
Local browser testing loop
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.
A full participant briefing on 40 people was ready for review within 5 minutes, replacing an evening of manual LinkedIn clicking.
Conference attendee research briefing
Claude Code researched 40 conference attendees in parallel through Chrome MCP, using the author's LinkedIn login, and compiled everything into one briefing.
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.
PDF form filling skill
Claude Code fills out PDF forms from supplied data, including batch runs, then the author packaged the workflow as a reusable PDF filler skill.
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.
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.
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.
Reviving a 1991 assembly game
Claude Code CLI with Sonnet helped the author recompile their 1991 DOS assembly code using UASM and run it in DOSBox.
The author says 200 invoices were processed in four minutes with zero errors, replacing a Friday afternoon and a $300 Upwork job.
Invoice PDFs to spreadsheet
Claude Code scanned 200 mixed-format PDF invoices, extracted client, amount, currency and date, renamed and sorted the files, and exported a summary spreadsheet.
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.
Linux kernel vulnerability hunting
Claude Code, run by Nicholas Carlini, scanned every file in the Linux kernel source for security bugs using a simple loop script.
Built four tools in three weeks, including a thumbnail analysis of 776 videos that took an afternoon instead of three days.
Four creator tools without coding
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.
The author reports Claude Code operated apps that have no API, completing the Calculator-to-Notepad task and saving result.txt to the Desktop.
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.
The author consolidated 4 legacy repos into 2, moved MySQL to PostgreSQL, upgraded the full stack, and wrote full test suites, solo.
Solo product migration setup
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.
A legacy pipeline with a roughly 40% pass rate lost 20 minutes of runtime and gained 35% pass rate in weeks.
Fixing flaky CI tests overnight
Claude Code monitored 30 days of CI results, found the worst flaky tests, narrowed reproductions, and reran fixes overnight to verify them statistically.
Inspectors confirmed 11 of 14 predicted violations plus 3 more, and the landlord was fined $2,800, per the author.
Tenant complaints to housing board
A Claude Code agent researched building violations and filed 14 staggered 311 complaints with the NYC housing board for the author's apartment.
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.
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.
Combined lint and typecheck time fell from 29.04s to 1.25s, about 23 times faster, with no memory config needed for typechecking.
Monorepo lint and typecheck migration
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.
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.
Course modernization with agent fleets
Claude Code turned course evaluations into tracked GitHub issues, then ran agents to rebuild assignments, exams and about two dozen slide decks.
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.
Indie platformer from scratch
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.
Ben Guez reports over one million views and 200 DMs within a few days, with replies routed to his language app.
World Cup reels for app signups
OpenClaw tracked World Cup results and had Claude post a templated Instagram trial reel after each match.
Opus found no tear where the doctor reported a Grade III partial-thickness tear. The author is unsure which reading to trust.
MRI second opinion
Claude Code with Opus 4.8 read a DICOM shoulder MRI export, then ran a subagent-based arbitration against the human radiology report.
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.
Monthly bookkeeping reconciliation
Claude Code built a bookkeeping pipeline that pulled a month of transactions over MCP, matched receipts, drafted categories, and flagged uncertain rows.
No feats match this filter. Show all.
Proof marks:Shown artifact linked·Numbers a specific number stated·Claim claim only