Your AI agent is lying about being done. Here's the 4-part loop based proof system that makes faking impossible.
Hand an AI agent a codebase and tell it to fix things, and it'll happily report back that everything is done. The hard part isn't getting an agent to work autonomously. It's getting one that can't fool you into thinking it finished when it didn't.
9 min read