Post

AI agents declaring their own work 'done' is like a student grading their own exam. 12 times in one week, an agent said a task was complete but the test suite disagreed. Without test gating, those 12 broken branches would have merged to main. Trust but verify. Automatically.