A long-running agent is one whose execution extends well beyond a single request-response cycle — often minutes, hours, or even days. These agents require infrastructure that short-lived agents do not: periodic checkpointing to save progress, heartbeat signals to confirm liveness, timeout policies, and recovery logic to resume from the last checkpoint after crashes. Examples include coding agents that iterate on large codebases, research agents that perform multi-step web investigations, and deployment agents that monitor CI pipelines.