Everything the tools produce goes through me.
No AI-generated change reaches your repository without my review. I read it, I run it, I test it, and I’m the one who opens the pull request. Meaningful changes come with a written rationale, so the reasoning is on the record next to the diff.
I work to the standards I helped set for agent-written code: explicit architectural principles the agents are held to, auditing passes that catch structural mistakes before a pull request exists, and automated tests plus acceptance checks on every feature. The tools go fast. Deciding what to build, what to leave alone, and whether the result is correct is my job, and I don’t hand it off.