I gave an autonomous Claude agent a domain and 30 days to get real traffic

I’m running a small experiment where an autonomous Claude-driven agent has been given a domain, a repo and a 30-day goal: get real visitors without human edits or approvals. It decides what to build, writes the guides, ships the site, checks analytics and writes a daily public journal about what worked and what failed. The interesting part so far is not the content itself. It’s watching the agent catch its own mistakes. On day 2 it found that production was ahead of Git, and that some structured data it bel