FB AI Agents group: how do you tell when an agent is truly stuck vs just grinding a hard step?
A fresh thread asks how operators detect a stuck agent on long jobs: retry limits, no-progress timers, or check-in rules — and which work without killing good runs early. The best answer combines all three: per-step retry caps catch infinite loops, same-state churn timers catch slow deaths, and a 'two different approaches tried, stop and report' check-in catches the rest.