What is Scheduled Agent?
Also called cron agent, recurring agent.
A scheduled agent runs on a time-based trigger such as an hourly, daily, or weekly schedule, rather than in response to a request. Nobody is present when it starts, so it must determine its own context, decide whether there is anything to do, and route its output somewhere a person will see it later.
Working unattended changes the requirements. There is no one to answer a clarifying question, so ambiguity must resolve to doing nothing rather than guessing. Output has no immediate reader, so it must go to a durable destination. And because the same run repeats indefinitely, a small recurring error compounds quietly in a way an interactive mistake does not.
Two mechanical issues dominate. Runs must be idempotent, so that a retry after a partial failure does not send the same message or create the same record twice. And overlapping executions must be prevented, because a run that takes longer than its interval will otherwise stack up against the previous one. Both are solved with run locks and recorded completion markers.
Scheduled work also fails silently. An interactive agent that breaks is noticed immediately; a nightly one can fail for weeks unnoticed. Standard practice is to alert on the absence of a successful run rather than only on errors. Cost deserves the same attention, since a frequent schedule multiplied by a long run is a recurring charge that nobody re-examines.
Schedules suit work defined by time, such as a weekly summary or a daily reconciliation. Work defined by something happening is usually better served by an event-driven trigger, which reacts immediately instead of waiting for the next interval. Many systems use both, with events handling responsiveness and a schedule acting as a safety net that catches anything missed.
Key points
- Starts on a timer with no person present to clarify anything.
- Runs must be idempotent and prevented from overlapping.
- Alert on missing successful runs, not only on errors.
- Time-based work fits schedules; reactive work fits events.
In practice
A pipeline review agent runs every Monday at seven. It pulls deals that have not moved in fourteen days, checks each for a recent email or meeting, drafts a short note on the three most valuable stalled ones, and posts a summary to a shared channel. If it finds nothing stalled it posts a single line saying so, which also proves it ran.