Simon Willison: verifying coding-agent output takes more than line-by-line code review

Simon Willison argues reviewing every line an AI coding agent writes isn't the best way to verify it did the job right.