Autopilot: let the AI send when it's confident
Autopilot lets PilotPM send replies with no human in the loop — but only on the channels you choose, only when it's confident enough, and never for anything sensitive.
The control center
Go to AI Intelligence → Automation. Every channel is a row with three settings:
- Off — no AI drafting.
- Draft — the AI drafts, a human always sends. (This is the default for every channel.)
- Auto — the AI can send on its own when confident.
For each channel you set a confidence threshold — how sure the AI must be before it may auto-send. PilotPM shows a data-grounded recommended threshold from your own history, with one click to use it.
Safety, always on
- Refunds, cancellations, password/login, money, and "I want to talk to a human" messages never auto-send — they always go to a person.
- Public channels only unlock auto-send after 100+ approved drafts, so accuracy is proven first.
- Everything is default-off, and you can flip any channel back to Draft instantly.
How it improves
PilotPM watches which drafts your team approves versus edits and gets better over time — a model that learns from your team's edits. Start on Draft, watch the recommendations, and turn on Auto channel-by-channel as you build trust.
Was this article helpful?
Or open the chat in the corner to ask follow-up questions.