Sebastian Bergmann, creator of PHPUnit, used an AI coding agent to implement the ACPATH metric (a software complexity measure) in 15 minutes. While the code looked clean, tests passed, and simple cases verified correctly, he found himself unable to confirm correctness for non-trivial cases without deeply understanding the underlying academic paper. The core insight: AI agents can generate code in a domain faster than a developer can understand that domain, creating a dangerous asymmetry. Code that looks correct and passes tests is not the same as code that is correct. The real risk of AI-assisted coding isn't bad code — it's plausible-looking code that stops us from questioning it.