I let an AI agent loose on my codebase. It tried to read my .env file in 30 seconds.

Not a horror story. Well. Kind of. A few months ago, Çınar and I were building a side project. Nothing fancy. Just two guys, a codebase, and way too much coffee. We started using Claude Code to speed things up. And honestly? It was great. It was writing code faster than we could review it, jumping between files, running commands, doing things we didn't even ask it to do yet. That last part should have been a red flag. One evening I left it running while I went to grab food. Came back. Looked at the terminal. It had read the .env file. Not because it was malicious. Not because someone hacked it. Just because it could. Nobody told it not to. There was no rule. No policy. No wall. It saw a file. It read the file. That's it. And I sat there thinking: this thing has access to everything. Every file. Every command. Every API call. And I have absolutely no idea what it's been doing for the last 20 minutes. No logs. No audit trail. No "hey, are you sure about this?" Just vibes. That was the mo