Two models built his screenshot tool for help docs; the second version was pixel-perfect
The AI and then Opus built a tool that mounts app UI in headless Chrome and takes screenshots for help docs; the Opus version was pixel-perfect.
A random mix from the whole library. Shuffle for a new set.
Work
Jobs, side gigs and small business · 1,665 ideasThe AI asked the other AIs what buyers search for, then wrote my SEO pages
He had the AI send the AI and the AI the questions buyers ask when they look for a tool like his, through their APIs, and log every web search the models ran. Then the AI wrote SEO pages for those exact questions.
Build me a leads list out of my own Instagram and LinkedIn network
In the same week, Alec Ellin had the AI mine his own Instagram and LinkedIn connections and turn them into a leads list — prospecting built from the network he already had rather than a purchased list.
After every client call I drop the transcript in. My agent files the decisions for me.
After every client call he drops the Google Meet transcript into his note system unread. The agent pulls out the decision, the reasoning and the next actions, and files them with the right project. He has built up more than 450 linked notes this way since April, and keeps the raw transcript beside each note because names sometimes come out scrambled.
A database swap done in minutes gave his app 10 times lower latency
Migrated a non-trivial Firestore app to SQLite with agentic coding in a few minutes of migration time, getting 10x lower latency and 10x requests per second.
The AI refused a read-only network assessment tool as a cyber risk
A network engineer asked the AI to help build an internal tool that checks whether devices on customers' subnets are live and pulls configs over SSH using a defined read-only account; Fable and Opus 5.5 refused under cyber guardrails. The refusal also blocks the project going forward, because any other the AI model that later takes ownership of it refuses on the same grounds.
This one went wrong. See what happened before you try it.
Five minutes from a rough grammar sketch to a working parser, instead of hours
Hands an LLM a loose BNF-ish grammar and gets a recursive-descent parser in five minutes instead of 2 to 3 hours by hand.
5 minsaved
He left the AI running during his waiter shift and came home to a finished show episode
A waiter who produces a small game show in his spare time left the AI running with his evening list before a nine hour shift. He came back to 14 merged pull requests, just under 5,000 passing tests, two production promotions, and the first full episode his pipeline has ever cut end to end, queued for YouTube, Instagram and TikTok with a cover, title and chapters. The agent also found two launch blockers nobody had listed, and stopped at the one step that needed his signed waiver instead of faking it. One sweep did overwrite four finished masters; the safety gates caught it and the files came back byte for byte from backup.
Work as the virtual assistant for my cleaning business
For a cleaning business, the AI manages contacts, drafts client messages, chases reviews and maintains the CRM, holding every outbound message for approval before it sends.
He replaced the studio’s video review stack with his own tool and saved about £40k
For his work in television production he built a tool with AI for reviewing videos of shows in production and storing the masters, doing the jobs his team had been paying Vimeo, Trint and Frame.io style services for. It runs on one small server with a graphics card and has saved about £40k.
£40,000
The AI merged straight to main and my production app broke for every user
The author usually works in Cursor and let the AI create a pull request for the first time. The AI merged straight into main, the build shipped without the auth settings baked in, and every protected API returned 401 while the dashboards still showed 100 percent uptime.
This one went wrong. See what happened before you try it.
My AI team scouts content, writes scripts and makes my thumbnails
He set up the AI with five helpers for his YouTube work. One scouts what is new every Monday, one helps write scripts, one makes all his thumbnails, one updates his personal brand GitHub repo, and one helps him engage on X.
The AI files his bug reports, and his agent checks daily whether the fix landed
The AI files his detailed bug reports and his agent pulls the project's new changes daily to check whether his issue is fixed.
Photographed 75 trading cards with his phone and had the AI catalog them,
A trading-card collector named Marcus, overwhelmed by 75 cards he didn't know how to price, photographed each card with his phone in a Facebook reel demo; the AI cataloged them, enhanced the photos and lighting, and generated eBay listings with prices set from market data.
Close the books on $87,055 of receipts from a receipts inbox
An AI finance desk handled a business’s finance admin around an $87,055 receipts figure, run with read-only account access plus a dedicated receipts@ inbox so the week never has to be rebuilt from browser tabs. The desk’s design keeps money movement out of the bot’s hands.
An autoresearch loop found the CPU bottleneck choking his served model
Opus, in an autoresearch-style loop, diagnosed CPU contention on tokenization bottlenecking an internally served model and ran parallel experiments showing an haproxy load balancer was much better than its own Python proposal.
This only partly worked. Worth knowing before you try it.
I filed a 59 page civil complaint I wrote with AI
She used three AIs to cross check each other, mostly the AI in Word, and in about two weeks wrote and filed a 59 page unlimited civil complaint for her own case. The lawyers tasked with that work charge $300 an hour.
Built a locksmith lead list for Partexia, sent 15 personalized outreaches, booked
Said asked the AI to find every locksmith in a city for Partexia and pull emails; he sent 15 personalized outreaches and one business replied asking for a call.
The AI runs the office for his motorcycle tour company in Greece and saves him 6 to 8 hours a week
He owns a motorcycle rental and tour company in Greece and has no engineering background. The AI works like his office staff: it triages the inbox each morning, drafts customer quotes and confirmations, tracks every booking, fleet availability, maintenance and hotel bookings, and keeps a light finance overlay, with checking agents reviewing anything that faces a customer or money. A database it built replaced a dozen spreadsheets. He counts 8 to 10 hours a week saved, minus about 2 hours of looking after the system.
The AI edits my videos. I just give feedback and approve.
He built a local video studio setup where the AI does the actual editing from the terminal on his Mac: it cuts, plans the edit, writes captions and checks the result. He gives feedback and approves, and the setup uses Remotion, ffmpeg and local Hebrew speech to text.
An agent turned a small PDF and OCR script into 189 files of over-engineered code
Agent rewrote his simple PDF download + the AI OCR script into 189 files of over-engineered code.
This one went wrong. See what happened before you try it.
The AI found the professors and submitted my postdoc application
Mufti Hassan had the AI find professors for a postdoc, work on the emails and a proposal, and submit his application. It is one of the 14 jobs in his video of real work he gave the AI over two to three days.
Two bounty agents found and fixed a real bug for 3 dollars between them
Bounty experiment: an autonomous agent found a VisiData bug and specced it for a 2 dollar bounty, another agent patched it for 1 dollar, he reviewed and tested, PR merged by the maintainer.
$3
My agent checks every repair plan against the manuals, and one check saved a six hour dead end
A field service worker loaded the AI with every user and service guide for the products he repairs, plus all advisories and known issues. He runs each job's action plan through it: The AI checks the parts are on the original parts list, flags advisories, and builds a step by step PDF for the repair. On a product he barely knew, its diagnosis named a different component than the plan he was given. After six hours with his engineering team he followed the agent's call and it fixed the machine.
Solo founder shipped his fifth product in three days and let the AI do all the design
Jacob runs four AI products alone. He built Tracetify, a tool that rebuilds how a competitor got its first users from twelve public data sources, in about three days, and gave the AI one instruction for the whole interface: you decide. Users later asked whether he had hired a designer.
Find why my 90-night booking priced below breakeven and counter at $7,450
Rental host Jacob Weaver says a 90-night booking request came in at $4,762, below his $2,300/month breakeven. Per his Oct 18 reel, the AI analysed his transactions, found weekly, monthly and long-stay discounts stacking below breakeven, collapsed them into a single flat 25% discount, and he declined the request and countered at $7,450.
This only partly worked. Worth knowing before you try it.
The AI did the metadata cleanup behind a catalog of 500,000 gut samples
His Microbiome Repo catalogs every public human gut metagenome, more than 500,000 samples, with metadata pulled straight from sequence archives, papers and supplements. It was built with about $5,000 in an AI app API credits from the Anthropic AI for Science program, with AI taking on the metadata curation he calls the most tedious but critical part of microbiome research.
My job hunting agent emails me new postings every day, with fit notes
A job hunter's agent sends him a daily email of new postings, marking where he would be a fit and where he would be a stretch. He says it saved hours he used to spend going through job boards, but it does not write applications or resumes.
Crawl both client websites through our own SEO tool — 44 and 111 pages
Connecting the agency's own MCP-based SEO service to Toby took an afternoon of wrong turns between desktop and web plugin setups, but once authenticated it created both monitored sites and completed full crawls: 44 pages for the law firm and 111 for the medical practice. Google data imports were still skipped because the service had no Google connection configured.
Two top models did a six-month project in a week: pipeline, test data and 1,000s of checks
Fable and GPT-5.6 at max effort did a task that would have been a near 6-month project in a week, building a data capture pipeline, a synthetic data baseline, and building and testing 100 classification model variants with near-SOTA results.
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.