CodingX · 2026-10
A user turned his 15K saved Wispr Flow dictations into a test set for the local Whisper model behind his Hermes Agent. The agent turned his Wispr dictionary into a hint list, diffed raw vs corrected text to find words Whisper keeps mishearing, made find-and-fix rules, and tested every rule against the full history. With no fine-tuning, dictionary words went from 99/240 to 183/240 on Wispr clips and 19/45 to 43/45 on sentences he read aloud. It now archives the data nightly and looks for new fixes weekly.
- Level Advanced
- Setup1–3 hrs*
- Evidence Strong evidence
CodingX · 2026-02
The agent sets up an ElevenLabs voice agent connected to its own gateway and adds a talk widget to the website.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingReddit
A non-engineer built a 'boring' vertical B2B SaaS with an orchestrator chat plus worker chats. Two paying clients and 6 leads, about $800 in credits spent.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
Sponsored
A native, clearly labelled card in the directory results.
Become the featured sponsor →
CodingReddit
Shipped in about a month on agent-friendly tooling (Next.js, Convex, Clerk, Stripe, Vercel). It was free for 3 months, and growth came entirely from long-tail SEO landing pages.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingX · 2026-10
Asked to merge the pull requests waiting on him, the agent merged seven, found one with conflicts, had a second agent rebase it, then merged that one too.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingWeb
Linus Torvalds posted a page-cache patch series and credited Claude Code with the code, the test programs and the review. Small-read throughput on his laptop went from about 10 million to 33 million reads per second.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingWeb
An MCP server plus Blockbench plugin lets Claude build models, unwrap UVs, paint textures, add bones and animations, check its work with screenshots and export. Every step lands in Blockbench's undo history.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingWeb
A developer runs a Claude Code session per repository; instead of editing each other's code they send messages. Over three and a half weeks, 81 sessions across 13 projects sent 624 messages, with up to nine active in one hour.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingWeb
An open-source Chrome extension records you doing a task once, a vision model turns it into a plan with variables and a trigger, and a builder outputs n8n nodes. Steps without an API become marked manual steps.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingWeb
Cockroach Labs ran a coding-agent pipeline with hospital roles (triage nurse, fellow, review attending, discharge nurse) and mandatory planning and review before merge. Over 5 months it processed over a million lines of code with only 7 reverts.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingWeb
Over 3 months and 500+ billion tokens, a developer and community members had AI agents decompile a popular first-person shooter into readable C++ that compiles, and wrote up how they set up the harness and orchestration.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingX · 2026-10
Asked for an update to a chess game, Grok Bot built “Halloween Chess” with witches, ghosts, skeleton horses, a giant pumpkin boss, thunder effects, and an AI coach—reported as fully playable, with a demo video.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingX · 2026-10
Claude couldn't find a half-finished session from August; Grok Bot found it in 12 seconds and put it in a folder so Claude could finish the work.
- Level Beginner
- Setup~30 min*
- Evidence Some evidence
CodingX · 2026-10
A user had Grok Bot monitor an intensive Claude Cowork task. It worked, but the coordinator used 7% of their Ultra usage in under 2 hours vs 2% for the agent doing the work.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingWeb · 2026-02
A café and coffee-roaster owner in Australia with little coding background used Replit Agent to build AssetShark, an asset-tracking and depreciation app that uses the Australian Taxation Office's rates. It has a React front end, a Node back end with login, a content system, an admin dashboard and a free calculator that brings in leads.
- Level Advanced
- Setup1–3 hrs*
- Evidence Some evidence
CodingX · 2026-09
Codex stopped responding on a user's home PC while he was out. From his phone, he asked his Grok Bot what was wrong. It checked the PC remotely and found Codex still running, with 539 threads, hundreds of leftover node processes and his 32GB of RAM nearly full.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingWeb
CLI-Anything wraps desktop software in a command line coding agents can operate (GIMP, Blender, LibreOffice, Calibre, and community tools). Install via CLI-Hub, or generate a CLI for an app that doesn't have one yet — works with Claude Code, Cursor, Codex and similar.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingX · 2026-10
After his kids asked why a straw looks bent in water, a parent asked Grok Bot to explain light refraction. It built a playable 3D game in a single HTML file: an archerfish shoots insects off leaves, first misses by aiming where the fly appears, then corrects for the bend in the light (n = 1.33) and hits.
- Level Beginner
- Setup~30 min*
- Evidence Some evidence
CodingX · 2026-10
Inspired by a developer who fit Minecraft inside a tweet, a user gave Grok Bot the link and asked for Snake that works the same way. The result runs on one Node server with a shared 72 by 72 world, where everyone who opens the post plays in the same arena, dead snakes turn into food, and bots fill the empty slots when fewer than 4 people are online.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingWeb
Co-op Anything is an open-source playbook and MCP server: point your coding agent at a game and it researches existing mods, reverse engineers what it must, tests with in-game bots and packages a one-click install.
- Level Advanced
- Setup1–3 hrs*
- Evidence Some evidence
CodingX · 2026-10
On day 7 of a 30-day test of Grok Bot for daily life, a user asked it to build a YouTube arcade game while he sat in the barber's chair. It studied live YouTube Playables games, picked a concept, wrote the game on its own computer, playtested it with a bot, hosted it and made the store covers. He walked out with a fresh cut and a playable one-tap game, Ring Drop.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Some evidence
CodingReddit · 2026-10
A Reddit user gave Claude Code one prompt: make the most realistic water scene it could in WebGPU, with a floating camera. It wrote the renderer from scratch with no engine or libraries (FFT waves, whitecaps, a physically based sky and sun glitter) in a single 61 KB HTML file. It screenshotted its own output and had separate subagents critique the screenshots so it wasn't judging its own work. The whole thing took about 30 minutes.
- Level Advanced
- Setup1–3 hrs*
- Evidence Some evidence
CodingX · 2026-02
Send a task over Telegram, the agent writes the code and opens a PR, and a nightly job has the agent review its own logs and improve.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Anecdotal
CodingReddit
A former trainer prototyped in Lovable and finished in Cursor. AI workout plans for coaches reached $2,320 MRR, but Claude API costs are $850 a month.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Anecdotal
CodingReddit
A solo founder used Cursor with Next.js, Supabase and Vercel. Reddit posts brought about 200K views, 4K visits, 385 signups and roughly $80 MRR.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Anecdotal
CodingReddit
A solo dev shipped a full-stack app with 3 LLM integrations. Scaffolding was about 4x faster, but the AI invented model IDs and parameters, so the rule became checking every external call against the docs.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Anecdotal
CodingWeb
An always-on agent with SSH, kubectl and 1Password checks health, analyzes logs, fixes pods and configs, deploys through infrastructure-as-code, and runs a nightly brainstorm.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Anecdotal
CodingX · 2025-08
Gave Perplexity Comet step-by-step manual test cases for a live site. It ran each one and reported results.
- Level Beginner
- Setup~30 min*
- Evidence Anecdotal
CodingWeb
A developer running Claude Code and Codex on the same codebase stopped them clobbering each other with one shared instruction file, a git worktree per agent, and a handoff file instead of re-explaining context.
- Level Beginner
- Setup~30 min*
- Evidence Anecdotal
CodingWeb
Every Monday at 7:00 a workflow checks a list of client domains and emails a warning for any certificate expiring within 30 days, so you don't hear about it from a customer.
- Level Beginner
- Setup~30 min*
- Evidence Anecdotal
CodingWeb
A macOS command-line tool plus a Claude Code/Codex skill lets the agent draw an arrow and a sign on top of any window to show you where to click. Clicks pass through and the arrow removes itself.
- Level Beginner
- Setup~30 min*
- Evidence Anecdotal
CodingWeb
One Claude Code session writes briefs, checks claims and reads diffs while separate sessions do the work, with an external task board holding state so nothing is lost when context compacts. Every agent report is re-run before it's believed.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Anecdotal
CodingX · 2026-10
Rusty Moore shared the exact Grok Bot routine a friend uses: every day at 6pm it opens the Cursor spending dashboard read-only, reports weekly usage %, reset timing, whether on-demand spending is still off, and an activity estimate — then screenshots and appends a log row.
- Level Beginner
- Setup~30 min*
- Evidence Anecdotal
CodingX · 2026-08
A user frustrated by slow website access found Grok Bot will connect to their own custom MCP servers just by being given the URL, beyond the official connectors.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Anecdotal
CodingWeb
Nubank used Devin to break up an eight-year-old ETL monolith with millions of lines of code. The work was split into many small sub-tasks, and Devin was fine-tuned on earlier migrations that engineers had done by hand.
- Level Advanced
- Setup1–3 hrs*
- Evidence Anecdotal
CodingWeb
GitHub's blog shows assigning an issue directly to Copilot's coding agent. It works in a GitHub Actions environment, commits changes, runs your tests and linters, and opens a pull request for review.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Anecdotal
CodingX · 2026-10
A full-time agent user has a Monday agent read the past week of Claude Code and Codex sessions (laptop + home server): repeated errors, blocked commands, corrections, expensive overnight failures — it proposes one concrete fix per pattern and waits for yes/no before changing anything.
- Level Advanced
- Setup1–3 hrs*
- Evidence Anecdotal
CodingX · 2026-10
A parent had Grok Bot build a witch's potion lab game: brew potions from a recipe in a bubbling 3D cauldron with seven ingredients, discover secret potions, pick Easy, Medium or Hard (Hard scores 1.5x) and turn on a kids mode with hints. Mix it wrong and it explodes.
- Level Beginner
- Setup~30 min*
- Evidence Anecdotal
CodingX · 2026-10
To let ChatGPT Dots communicate with Grok Bot, a user asked Grok Bot to set up a webhook, passed the link to his Dot, and that was it.
- Level Intermediate
- Setup1–3 hrs*
- Evidence Anecdotal