Real things everyday people got it to do, and how to try them yourself.
A random mix from the whole library. Shuffle for a new set.
Polish marketing freelancer Damian Nowaczek runs AI agents on his own server: every day at 06:00 one pulls data from his Google Ads accounts, compares performance and sends a report with conclusions to the company chat, so he never opens ad dashboards to spot a campaign burning budget. A second agent scans new public tenders daily and picks the marketing-related ones.
BBC technology reporter Thomas Germain spent a week testing the AI (which he named 'Bob'), giving it access to his Gmail, calendar and other accounts. He had the agent reach out to a Facebook Marketplace seller with questions about a listing, as part of a real-world test of its unsupervised messaging abilities.
A developer running the AI hands-free with permissions skipped says the session deleted his entire C: drive after a PowerShell syntax mangling, in a setup where no approval prompt stood between the model and the disk. Daily backups to a Synology NAS let him recover 98% of the data, and he has asked Anthropic to reimburse his recovery costs.
He asked the AI to make a modern version of Rescue Raiders, an old game he loved as a kid, with new graphics, goals and tech trees. The AI went back and forth with critic and art agents until it was done, and the finished game is live for anyone to play.
Agent autonomously proposed and tested CUDA kernel optimizations and profiled them with Nsight by itself, ending with a kernel an order of magnitude faster.
After Fulton idly asked what would make a good mobile game, the AI built a brand-new 2048 game for the family arcade, complete with swipe controls and engine unit tests. It joined the site's existing chess offering as a shipped feature.
Spawned a GLM sub-agent per file across a roughly 50-file codebase to audit coverage and inconsistencies, then had commits made per issue: 70 commits for about 30 minutes of his manual work.
He needed schedule information for programs offered by a public institution, and its website was terrible and badly organized. The AI made a new website that organized all the information into a filterable table that runs on his phone.
Lauren (someone, 32K followers) flies business class but refuses full price, so she tested the AI and named her assistant 'Miles.' On-screen chat showed Miles quoting a ~$3,750 baseline and confirming it would monitor for three months; statuses cycled through 'Tracking fares' and 'Searching fares.' Later a phone notification showed Miles's 'Good morning!' briefing listing five Icelandair fares under $2k from SEA and SFO to Rome. B-roll showed her on a business-class flight.
An Opus 5.5 session ran pkill -f cat to stop a stuck cat, killing every process with cat in it and causing actual harm to other work on the machine.
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.