Scheduled agent

An agent run started by a recurring schedule rather than by a new interactive request.

A scheduler can launch an agent at a stated time, supply the task definition and current state, and write results to a durable destination. Morning briefings, weekly metric summaries, and periodic log analysis are examples. The scheduler is the trigger, not the prompt: the run still needs instructions, sources, permissions, and an output contract. Local and cloud products differ in whether the computer must be online and how missed runs behave.

Builder example

Scheduled work needs explicit state, idempotency, time zones, retries, alerts, output validation, and missed-run behavior. A run can complete successfully while using stale data or writing to the wrong destination, so infrastructure success and task success need separate evidence.

Common confusion: A scheduled agent is still an agent, with all the reliability requirements that entails: error handling, escalation policies, and output validation. Putting an agent on a cron job without those guardrails creates a system that fails silently at 3 AM.