In May 2025 I tried to build thirty AI agents in thirty days, in public. The plan was six agents a week with a rest day, from a personal assistant to research, finance, travel, email and translation. Each agent got a spec first, and each day was a branch merged by pull request.
I got to day 22. The first week ran on schedule. After that each day took a little longer than a day, and by early June the challenge had stopped. Days 7 and 20 are only partly built, and days 23 to 30 were never started.
The history shows how much I was learning as I went. The stack shifted inside single days: one agent moved to a FastAPI and React app, and another was rewritten as a CLI. A few MCP servers came out of it as side projects, including beehive-mcp-server. Day 16 brought me back to Android for the first time in six years, with an offline voice assistant running whisper.cpp on the phone (VoiceAssistant). I wrote about where it fits in How I got here.