Enabling autopilot and the always-escalate safety list

Autopilot lets high-confidence AI drafts send automatically without a human click. It's off by default — you turn it on per workspace once you trust the AI's calibration.

How auto-send decides

A draft only auto-sends when all of these are true:

• Workspace auto-send is enabled.
• The draft's confidence clears the channel floor (email is 0.85; public channels like app-store reviews are higher).
• The model didn't flag itself uncertain.

The always-escalate safety list

Some topics are never auto-sent, no matter how confident the AI is. Any draft (or the customer's message) that mentions passwords, account lock/unlock, refunds, cancellations, money or percentages, contracts, or a request to "talk to a human" is routed to a person for review. This is intentional: these are high-stakes, account-security and financial topics where a wrong automated reply does real damage.

So even a perfect password-reset draft will still wait for a human — by design.

Getting started safely

Turn autopilot on for low-risk FAQs first, watch the AI Intelligence view for calibration, and expand from there.