AI agents for developer: 81 real examples
Agentic Feats lists 81 sourced examples of AI agents in developer work; 47 link the artifact; most common agent: Claude Code (26). Newest first.
- Feats
- 81
- Shown
- 4758%
- Agents
- 14
- Updated
- Oct 13
Agents used here14
Industries here5
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.
Self-hosted agentic software factory
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.
Claude gave failed builds for a week and used up the 20x Max weekly limit. Sol had worked 20+ hours without success.
VNC client build attempts
The author asked Claude, then ChatGPT 5.6 Sol, to build a bare-bones VNC viewer for Windows 10 that avoids dropped keystrokes.
Within 90 minutes Claude produced a functional proof of concept confirming an authentication bypass across four tested endpoints.
Reverse engineering a papercut zero-day
Claude Code analyzed a patch for an actively exploited Papercut NG vulnerability and worked to build a working exploit chain.
The best recorded run scored 39,359, reached 4,924 meters, and hit a combo of 117.
Browser game and PPO player
Grok built the browser game Roofline, then trained a PPO agent that plays the live page.
The fleet manages over 200 cloud agents at once, up from 15 managed by hand, and built Grok Bot iOS v0 in three weeks.
Engineer bots managing cloud agents
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.
On a payment service v2 to v4 test, it caught an outdated SDK, returned DO NOT SHIP YET, then SAFE TO SHIP in simulation.
Pre-deploy release gatekeeper 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.
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.
Million-line codebase refactor
Hermes Agent split a million-line Python repo into 36 groups and dispatched 1,393 subagents in git worktrees to refactor it.
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.
Faster Rust ML code via benchmarks
Coding agents wrote Rust ML algorithms like UMAP from scratch, then iteratively optimized them against criterion benchmarks.
Androidclaw traces broken builds to the relevant pull request and incident, and sometimes prepares and merges fixes itself.
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.
The unfinished output is work the author would have quoted at about €2.5k, built in one overnight run.
Marketing site from Figma
Devin AI connected to a Figma file through MCP and built a WordPress marketing site with easy content management overnight.
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.
Rule-citing AI dungeon master game
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.
Users resolved issues by talking to Devin, engineers only merged when needed, and Devin Review caught bugs.
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.
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
Claude pulled 60 documentation pages into local Markdown files in about 49 seconds, including roughly 10 seconds of backend cold start.
Docs crawler MCP server
Claude used the author's pulpie-mcp server to crawl a documentation site and save every page locally as clean Markdown files.
It cost about two months of Pro subscription over half a working day, and ended up slightly slower than doing the migration by hand.
Password reset migration experiment
Claude Code with Opus migrated a password reset flow in a legacy web app, with the author closely supervising under a pre-made plan.
The orchestrator produced a complete video with clips, music, and effects, shown in a linked video.
Agent-orchestrated AI video
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.
The author got a working Rust editor, .NET 10 scripting, multi-threaded runtime, culling, weather, lighting, water simulation and hot reloading within minutes.
Game engine from one prompt
Grok Bot ran four bots with one coordinator and built a Unity-style game engine from a single prompt, in minutes.
Db2 support shipped in under two days for a $4172 token bill, versus 9 months and about $160K for Oracle; 27 PRs merged.
Hospital-style agent pipeline ships Db2 support
Cockroach Labs ran a multi-role coding agent pipeline modeled on a teaching hospital to add Db2 support to its MOLT migration tool.
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.
Three agents build multiplayer Counter Strike
The Instant team gave Codex, Claude Opus 4.5 and Gemini 3 Pro the same prompts to build a 3D multiplayer Counter Strike clone.
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.
Claude Code plays RollerCoaster Tycoon
Ramp Labs forked OpenRCT2, built a kubectl-style CLI, and let Claude Code manage a theme park through it.
Word starts and draws its ribbon, OneNote signs in and saves edits, and a retail Home 2024 key activated through account sign-in.
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.
About 19k lines of Rust across 154 commits in three days. The engine passes 64 SQLLogicTest queries and 282 AI-written unit tests.
SQLite-like engine in Rust
Claude, Codex, and Gemini workers, two of each, built a SQLite-style database engine in Rust by claiming tasks through git lock files.
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.
Coding agent built on LLM
Claude Code on the web wrote a spec, then built and tested a coding agent library on LLM using red/green TDD.
A VS Code extension was built overnight with 91 of 91 tests passing and a PR opened, costing about $0.35.
Overnight builds and inbox triage
OpenClaw on a Raspberry Pi 5 triages email, builds code from a TODO list overnight, and sends daily research reports to a security consultant.
Out of 20 tasks, Devin had 14 failures, 3 successes, and 3 inconclusive results, with no pattern predicting which would work.
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.
Out of 20 tasks, Devin had 14 failures, 3 successes, and 3 inconclusive results, with no predictable pattern for which would work.
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.
The author upgraded kentcdodds.com across all dependency tiers, finishing a Remix v2 to React Router v7 migration with routing matching the original structure.
Dependency and framework upgrade
Cursor listed outdated packages by upgrade difficulty, then upgraded them in stages, including Remix v2 to React Router v7 on a 42k-line site.
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.
Multi-agent stock analysis pipeline
The author built a LangGraph system where valuation, sentiment and fundamental agents plus a coordinator rate four stocks and backtest the resulting portfolio.
Devin opened a PR in about four hours and got the build passing, but runtime issues remained and the author gave up.
Vue 2 to Vue 3 migration
Devin was asked to migrate a production Electron app from Vue 2 to Vue 3 with Composition API, script setup, and Pinia.
The author shipped a redesigned personal site, nearly right in one shot, after a few more iterations.
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.
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.
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.
A working prototype runs on the handset over mobile data: typed chat via DeepSeek, battery check, torch, calls and alarms, tested on the phone.
AI chat app on Nokia 110
Codex helped the author reverse-engineer a Nokia 110 4G's firmware and build a native AI chat app with a custom agent.
No feats match this filter. Show all.
Proof marks:Shown artifact linked·Numbers a specific number stated·Claim claim only