Fallback

A predefined alternative action an AI agent takes when its primary approach fails—such as escalating to a human when confidence is low, switching to a simpler model when latency spikes, or returning a canned response when the knowledge base has no match. Well-designed fallbacks prevent agents from failing silently or producing low-quality outputs, and they are the recovery layer of reliability engineering: a fallback is the *specific* alternative action, while graceful degradation is the design principle that there always be a safe one to drop to. Because agents are non-deterministic and retry often, any fallback that re-attempts an action must target an idempotent operation, or the retry itself becomes a new failure.