A sandboxed agent checks daily whether the product spec and the code still agree
Sandboxed agent checks spec and code drift daily, triages findings, and surfaces issues; it revealed product miscommunications on the team.
A random mix from the whole library. Shuffle for a new set.
Work
Jobs, side gigs and small business · 1,668 ideasBuild a SaaS in a weekend with AI — and catch the payment bug it approved itself
Ryan Mercer spent a weekend building a SaaS with AI and Next.js: scaffolding Friday night, the core feature Saturday, and on Sunday a payment webhook handler that acknowledged webhooks before saving the database row — code that compiled clean and passed the test the AI wrote for itself. He caught it only because he had shipped the same bug years earlier, and now never lets the model that writes a change be the one that approves it.
This only partly worked. Worth knowing before you try it.
Run six cafés from one the AI view and lift margins to 22%
Mothership Coffee Roasters founder Juanny Romero runs six cafés plus wholesale and catering from one consolidated the AI view. Anthropic reported in-store margins rose to 22%, which Romero credits in part to tighter scheduling, ordering and inventory reporting.
The AI found out my five new subscribers were bots
Five new newsletter subscribers looked like direct traffic, as if five people had typed his URL by hand. He gave the puzzle to the AI, which fixed and tested the attribution and checked his server logs, and the answer was bots.
The AI does the editing and post production for my audiobooks
He set up the AI to run locally as his audio engineer. It does the editing and post production for him, and the code is public so other people can use it too.
A daily Sonnet routine answers his routine email and blocks time for the rest
A daily routine on Sonnet reads his email, answers the routine messages from templates, and puts the rest on his todo list. When a task needs more than an hour, the routine blocks time for it on his calendar.
An overnight agent run produced 19 engine patches and up to 10x speed, stable for 3 weeks
Let Fable work overnight on llama.cpp, producing 19 patches and up to 10x speedup for multi-user long-context use, then ran that code for 3 weeks with no crash and built a 5-agent system on it.
Our AI assistant publishes our social posts every morning
A small affiliate business says its AI assistant starts every morning by reading click logs by source and checking journalist requests, then it publishes the next social posts. A person still handles identity checks, payments and signups.
Build websites on your own idea and sell them to local businesses
The user's the AI came up with the idea itself: it built websites and the user made the calls, closing two sales out of roughly 50 outreach attempts to local businesses.
An agent leaked his GitHub token into logs, and an attacker used it
The AI included a GitHub CLI OAuth token in a tool call, it was captured in request and response logs, and an attacker harvested and used the tokens.
This one went wrong. See what happened before you try it.
I made a Chinese lesson video with AI, the AI and a voice tool
He produced a Mandarin grammar lesson video from his own writing from a few years ago. The AI made the visuals with Remotion, the AI did the drawings and BytePlus TTS did the voices, and the finished lesson is up on YouTube.
His diagram library fork got a live demo, built by an LLM in his own terminal widgets
Forked go-mermaid and had an LLM add a live Mermaid and SVG demo using his terminal and WASM widgets, published online.
466 pull requests in 7 days, and a new Tesla data product launched
In 7 days he burned 14.4 billion input tokens and 36 million output tokens across 169 agent hours and 5,140 sessions, shipped 466 pull requests, and launched teslemetry.pro. A tool called Firstmate orchestrated the AI work.
The AI moved his product's database over a weekend; months later he is still happy
Migrated his product from DynamoDB to Postgres over a weekend with AI and has been happy with the change 4 to 6 months later.
Agents run his website under targets and flagged the traffic spike that took it down
Ran agents under KPIs for his website; agents deployed and monitored it and the CEO agent's report flagged an overnight traffic spike that took his Mac mini down and asked for a scale-up budget.
This only partly worked. Worth knowing before you try it.
Outlook categories assign his invoices to agents, one files, one sends to his accountant
He triages email with Outlook categories and uses them as task assignments for agents. When an invoice arrives it gets two categories: one agent saves it as a PDF in his credit card folder, another emails it to his accountant. Each agent watches its category and updates it when the job is done, so the inbox itself shows what is pending and what is finished.
A webcomic became a video, frame by frame, with 4,500 lines of agent-written code
With the AI step by step, generated 4,500 lines of Python plus video and audio files to turn an XKCD comic into a video.
He built his own YouTube studio with AI and it cuts his filming for him
A creator who could not afford an editor paid for the AI Max and built his own post production studio, about 31,000 lines of code. Its smart cutting takes his raw filming and builds a DaVinci Resolve timeline for him, saving about 95% of the cutting time, and it also generates shorts and uploads with titles, descriptions, chapters and tags.
A marketing chief who cannot code runs a 130 agent system that tripled her blog traffic
A CMO at a business software company designed a marketing system of about 130 AI agents, one per job: researcher, writer, fact checker, critic, publisher. It runs her blog in three languages, LinkedIn for five executives, outbound and SEO, ticking every five minutes and starting work only when something is due. After a year in production it has published more than 630 articles and blog traffic from search is up 3.9 times. Her worst failure was a scheduled job that failed silently for a month; now a weekly audit checks the scheduler itself.
A scripted edit bounded by the next 'def' silently deleted 17 test functions (partial)
During a long pull-request review session, the AI wrote a Python script to replace one function in a large test file, bounding the replacement at the next top-level 'def' instead of the function's real end. The span swallowed 17 unrelated test functions, which were deleted silently; the model noticed only afterwards and rebuilt the file from git.
This only partly worked. Worth knowing before you try it.
Every idea is a real post from someone who tried it, summarized in our words, with a link to the original. App names are swapped for plain words.