Jaipur — Okay so last Saturday was Fetch-A-Thon 3.0, the Fetch.ai Innovation Lab’s AI agent hackathon. One day, about seven hours of actual building, and a challenge that basically said: build an agent that does something, not just talks about it.
Madhura and I built ForgetMe. You give it your email inside ASI:One and it finds every data breach you’ve been part of, tells you which accounts actually matter, and then emails the companies a formal “please delete me” letter. Only after you say yes, obviously. It’s five agents in a trench coat: one finds the breaches, one judges them, one writes the letters, one chases the companies that ignore you, and one talks to you and pretends the other four don’t exist.
Things that went wrong, in order
The breach database we planned on wanted a 52 dollar yearly subscription, at 10:30 in the morning. We found a free one by 10:33. Python 3.14 decided agents shouldn’t be allowed to start at all. Then Stripe renamed two of its own settings on us mid-build, one after the other, like it knew. (I’d like it noted that every error message was extremely polite about it.)
Things that worked
All of it, by the end. Little cards inside the chat with checkboxes, a privacy score out of 100, a warning label for crypto accounts, a receipt for every deletion request, and a paid tier running on Stripe test mode. The agents now live on my JARVIS server, in their own little room, with a strict rule against touching JARVIS’s things.
The whole day was fuelled by Coffee Bites. Not coffee. Coffee Bites, the candy. Madhura and I went through an amount I am choosing not to count, on the grounds that the number would become evidence.
Result: not placed. I am not crying. (the gazette is wet and ink bleeds through.)
Honestly the best part was the first test run, my own email coming back with breaches from accounts I didn’t even remember making. Somewhere out there a website from my past still knows things about me. ForgetMe’s coming for it.
More when there’s more…