Agent tasks overview
What agent tasks are, the inbox model, and who can act on them.
Agent tasks are the items that appear in your inbox when an automation fires and requires your approval before executing its step.
Think of the task inbox as a supervised to-do list for the agent. The agent proposes steps (drafting outreach, advancing a deal stage, re-researching a client) and you review, approve, reject, or snooze them.
Task lifecycle
A task moves through these statuses:
| Status | Meaning |
|---|---|
| Pending | Awaiting human review |
| Approved | Operator approved; step has been executed |
| Rejected | Operator rejected; no action was taken |
| Auto-executed | Automation's autonomy was auto_execute_safe_steps; step ran without inbox review |
| Failed | Step was approved but encountered an error during execution |
| Snoozed | Operator deferred; task will reappear after the snooze period |
Who can act on tasks
Users with editor, admin, or owner roles can approve, reject, and snooze tasks. Read-only members can view the inbox but cannot act.
The relationship to automations
Every task is created by an automation. The task record includes a reference to the automation that created it, so you always know why the task was proposed.
Tasks created by an automation with auto_execute_safe_steps autonomy do not appear in the inbox. They execute immediately and are logged in the audit trail.
See also: inbox, approve, reject, and snooze, automations.