A Japanese learner who had used jisho.org for years hit a word its handwriting search could not match. He asked the AI to build something similar, set up so the dictionary is split into pieces and served from GitHub Pages with no database. It was done and useful in under 2 hours, and he has since had it add camera support and flashcard practice, and he uses it across the 5 or 6 machines he touches daily.
Runs a personal-assistant session via Signal that tracks tasks and delegates to per-project agent sessions on other machines; it is now his only direct interface.
The AI took the raw JSON vote files and, using the state's own rules for overvotes, skips and duplicate votes, built the full preference schedule for a US House race. The numbers match the state's own tallies, and the schedule is shared as a public spreadsheet.
Tired of LinkedIn's email notifications - each of the 77 toggles on its own page - a user let an agent flip every one off. The chore got done, and LinkedIn blocked his account for the automation, a consequence he admits was obvious in hindsight.
This one went wrong. See what happened before you try it.
His role was eliminated on Tuesday morning. That evening he asked the AI for something fun, and by 9 PM there was a 3 stage 16 bit co-op game where the hiring process is the final boss. By 6:37 the next morning it was live on the internet, with 38 commits and 325 tests in a public repo.
She has been building SurfAnnotate with AI and two colleagues, a free browser tool that runs on Neurodesk. It makes drawing regions on cortical surfaces a lot easier for researchers.
Had the AI reverse engineer 3D Space Cadet Pinball using Ghidra and radare2 and port it to Rust and Bevy; playable after 2 hours, plus an extra prompt for MIDI music.
His 3D Systems Sense 2 scanner had no working software left, since the maker and the other scanner apps had all dropped support. The AI updated the program he uses so modern RTX graphics cards work and found a data limit that capped his sensor at half its native resolution, and he later posted that it works.
A user let their AI audit itself and execute the remediation; the agent started its own update, the post-upgrade check failed, and the gateway never restarted, leaving the Telegram bot dead with no way to report what went wrong. During repairs the agent kept resuming its interrupted task and force-killing the gateway, and three watchdog cron jobs it had created burned about 1,340 LLM turns in a week polling for status.
This one went wrong. See what happened before you try it.