Michael JohnsonOpenAI announced Scheduled Automations for the Agents API on October 23, 2026. The feature lets developers run agent tasks on a cron schedule or at a specified time, and, when the automations tool is enabled, lets an agent schedule follow-up work. The announcement was published in the OpenAI developer community; implementation details are available in the Scheduled Automations documentation.
For applications, this adds persistent scheduling to an API where agents work through sessions that retain configuration and context. Teams can more easily assign recurring checks or delayed reminders to an agent without arranging each occurrence as a separate manual run.
How scheduling works
A schedule specifies the target session and when a task should be delivered. For recurring runs, the developer provides a cron expression and an IANA time zone, such as America/New_York. For a one-time run, the developer specifies a future Unix timestamp in seconds.
A task can create a new session for each run or continue an existing one. The first option suits independent recurring operations; the second preserves the session’s previous context. After creation, a schedule receives active status, but the task runs at its scheduled time: creating the schedule does not trigger it immediately.
Developers can also give the agent the automations tool. It can then schedule a follow-up check within the current session. This option is useful for multistep work where the next run depends on the previous result, such as periodically checking a task’s status.
Task delivery is separate from task execution
Each run creates an invocation—a record of an attempt to deliver the task to the target session. Developers can check the invocation status and session ID. If the session is busy, the task waits; the session receiving the task does not, by itself, mean the agent has finished its work. The session’s progress and result are tracked separately through session events, as described in the documentation on invocations and results.
In practice, this means monitoring should track two states: whether the scheduled task reached the session and whether the session completed it. For recurring processes, it is useful to store the schedule ID and connect it to the run history to identify missed runs and investigate failures.
What to check before implementation
- Time zone. Specify it explicitly and check how scheduled times align with daylight saving time changes in the selected zone.
- Session type. Use a new session for independent runs and an existing one when a task needs prior context.
- Queue and results. Account for delays in delivery when a session is busy, and remember that accepting a task does not mean the work is complete.
- Schedule management. The API supports creating, updating, pausing, resuming, and deleting schedules; check future runs after changing the configuration.
Scheduled Automations adds managed schedules to the Agents API, with each run linked to a session and tracked separately. When implementing the feature, developers need to decide when a run requires accumulated context and monitor both task delivery and its result.