Agent Durable Execution in Production: Preventing Duplicate Long-Running Tasks with Checkpoints, Idempotent Side Effects, and Replay Contracts
Long-running agents face their biggest risk during recovery: duplicate tool calls, duplicate writes, and duplicate messages. This guide covers checkpoint boundaries, idempotent side effects, deterministic replay, and safe recovery using LangGraph, Temporal, and Microsoft Agent Framework, with practical operation_id, side-effect ledger, and state machine designs.