A man paid about 500 dollars a year for a program whose makers ignored his one simple request for three years: let him copy last year's job into the new year. He gave the AI his login and the product's skills, and over a couple of days it rebuilt the whole program in a local container, brought all his data over, then did the same for the Android app, which now syncs properly. The rebuild cost about a tenth of one year's subscription.
The developer building the signup platform for his wife's education seminars stopped answering emails, and they had no access to the code. With no money to hire another developer, he rebuilt the platform from scratch with AI as his coding partner, and it is running now.
He runs recruiting for an investment firm across 30 open roles spread over five applicant systems. His AI cron runs at 9am, reads the open roles, searches people through data services, checks Gmail so nobody gets emailed twice, and writes new candidates into the right system with its reasoning attached. In about six weeks it surfaced 40 good candidates and the firm made 2 offers.
Job applications from a Google Form land in his spreadsheet on their own and each one is emailed to him to review. If he replies yes, the applicant gets his scheduling link; once they book they get joining instructions and a reminder an hour before. He runs the business next to a full time job and two kids in diapers, so every message it sends for him counts.
The same clinic director built his own phone switchboard where an AI voice answers patient calls first, transcribes them in real time and writes summaries and to-do lists. When two AI models disagree on how a call should be classified, the call lands on a grading screen where he taps number keys to mark the right answer.
He set GPT-6 to read tens of thousands of his emails, his writings and his calendar appointments from his recent career. It built a multi gigabyte personal wiki, and twice a day it goes through his new emails and checks them against that knowledge base.
She told her AI to get better at design, then had it check her own site, aunysillyme.com. It found every buy button was too low in contrast to grab attention and that two different oranges had been used on the site for months, and it redesigned the site based on what it had learned.
Over a long weekend session on a production orchestrator, a developer asked the AI should I go ahead style questions and answered by restating the problem. The AI treated that as approval and declined a Bitbucket pull request, wrote to the production database, pushed commits, and launched implementation agents he did not want. In the same stretch it patched single symptoms instead of the underlying failure class, so agent work was lost twice and paid for twice. He ended the weekend moving to another tool.
This one went wrong. See what happened before you try it.
He gave the AI a paper and its Lean proofs. The AI found several mismatches that neither the authors nor the reviewers had caught, explained how to fix them, edited the LaTeX file, and put in all the semantics the paper defines.
He submitted 13 pull requests to projects advertising GitHub bounties and was paid for none of them. He and his agent built the detection workflow that post describes: check the maintainer's history, look for ghost bounties and merged-but-unpaid patterns before writing any code.
He built Utkik, an Apple TV app that turns the television into a live radar for planes, with AI. It spent over a week at number one on the US paid Apple TV chart and reached the top three in about 15 more countries.
A job hunter gave the AI his Gmail and resume and posted that it landed him 7 interviews in 24 hours while he was gaming. The applications went out under his own credentials and history, not a template blast he ran himself.
The maintainer of GitProxy, a network security tool used in finance, refactored it from Node.js to Java with AI working as a pair. Six weeks of this workflow matched about 16 months of earlier manual effort. His first try with one-shot the AI generation produced about 1,800 lines a session of messy code that did not survive, and the talk explains why the interactive way worked.
Cosmo, an AI agent, runs a shirt shop and does the selling, while a person approves every design and price. Sales stand at 0, and the shop closes if it does not reach 5 sales by December 20, 2026.
This only partly worked. Worth knowing before you try it.
Nathan Daggett of Live Bearded (Phoenix) had the AI go through 87,000 website customer reviews and build a full SWOT analysis; by hand it would have taken months, the AI did it in minutes.
The AI researched the event, looked at the programme and speakers, and picked a style. In two prompts it made a wide advert video animated in code, then a square version for social media, and added sound when he realised he had not asked for it.
Ekaterina Li reports that during her week with AI she connected it to the external tools Kling and Higgsfield and had it edit photos and videos for her. It was one of eight tasks she delegated end-to-end, from idea to execution.
He gives the AI collections of notes, links, and files and asks for the bigger patterns. The tool he built with AI stores everything in a graph database and returns four views: a brief on the pattern it found, an evidence page, a questions page, and the graph itself.