Task Autonomy defines how much the assistant may do when handling new task work. You need access to AI Assistant settings and explicit operational approval before changing autonomy, channels, or limits. Merely opening this page does not authorize research runs or outbound messages.

What you’ll see

The Task Autonomy page, where you choose how much the AI assistant does on its own, then select Save Changes.

Review the policy

  1. Open Settings → AI Assistant → Task Autonomy.
  2. Expand How much the assistant does on its own and inspect the selected level.
  3. Review What the assistant may use, daily message caps, and Quiet hours.
  4. Compare the policy with the team's approved workflow and configured sender profiles or numbers.

Safe researches public information and writes internal notes without drafting or sending messages. Draft and wait prepares messages for a person's approval. Autonomous can carry out permitted work within the configured controls. These levels describe different operating permissions, not merely different writing styles.

Available capabilities include Public research, Notes for your team, Email, Text messages, and WhatsApp. Sending still depends on channel setup, consent, limits, and quiet hours. A policy setting is not evidence that any specific message was sent or delivered.

Save only the approved policy

Selections remain in the policy form until Save Changes. Review all capabilities and caps together before saving; enabling autonomy with an unintended channel can alter future work. The page states that changes apply to new work only, so this is not a documented way to cancel work already underway.

Verify and recover

After an authorized save, reload and inspect the policy. If a save error appears, resolve it before assuming the new limits are active. The reviewed load handler can fall back to default policy values when loading fails, so displayed defaults alone are not proof of the stored policy.

For uncertain or already-started automation, inspect its history and use the appropriate operator workflow rather than repeatedly changing autonomy.

Related: Automation Health, Tasks, Communication.

Screenshots

The screenshot for this guide is shown above, under “What you’ll see”.