Agent loops generate new reasoning each run. State machines execute the same trace every time. For regulated AI, only one of ...