How would you keep an agent from looping on the same failed action?

Example Answer

I would make retry behavior explicit instead of leaving it to the model’s instincts. The system should track attempted actions, failure reasons, and retry count, then use policy rules to decide whether to retry, change strategy, escalate, or stop.

I also want failures classified. A timeout, invalid argument, permission error, and contradictory tool response should not all trigger the same recovery behavior. Agents loop when the system treats every failure like generic uncertainty.

A good agent can recover, but it should recover with state and memory of what already failed. Otherwise retries are just expensive repetition.

What I always try to avoid is giving a process answer that sounds clean in theory but falls apart once the data, users, or production constraints get messy.

Common Poor Answer

A weak answer is saying you would add a prompt telling the agent not to loop. Loop prevention belongs in orchestration logic, not just wording.