# AI Autopilot

> Let the agent answer on its own, and take back a single conversation without stopping it everywhere else.

- Source: https://support.unifunl.io/en/help/inbox/ai-autopilot
- Category: Inbox (`inbox`)
- Language: en
- Updated: 2026-08-20

---

AI Autopilot is the switch that decides whether your agent replies by itself in a given conversation. It is the difference between an agent that assists you and an agent that works for you.

## Two switches, two scopes

Autopilot is easy to confuse with the workspace-wide **Active** switch. They do different jobs:

| Switch | Where it lives | What it controls |
| --- | --- | --- |
| **Active** | Top of the AI Center | Global. Turning it off stops the agent across the whole workspace. |
| **AI Autopilot** | In each conversation's bar | Local. The agent answers on its own in this conversation only. |

![The AI Autopilot switch in a conversation's bar.](https://support.unifunl.io/help/screenshots/21-autopilot-onoff.png)

![The global Active switch at the top of the AI Center.](https://support.unifunl.io/help/screenshots/22-agent-active.png)

## Take over one conversation

1. Open the conversation in the inbox.
2. Turn **AI Autopilot** off in the conversation bar.
3. Reply yourself. The [writing assistant](https://support.unifunl.io/en/help/inbox/writing-assistant) can still draft for you.

The agent keeps answering every other conversation. Turn Autopilot back on when you are done and it resumes on this one.

> **Good to know** — turning Autopilot off applies to that conversation for everyone in the workspace, not just for you. Agree with your team who is taking over.

## When the agent stops on its own

You are not the only one who can pause the agent. It steps back by itself when a conversation is [escalated](https://support.unifunl.io/en/help/inbox/escalation) — when a customer is clearly unhappy or explicitly asks for a human.

## FAQ and troubleshooting

### The agent is not replying even though Autopilot is on

Check the global **Active** switch in the AI Center first — if that is off, no conversation gets an automatic reply. Then check whether the conversation carries an escalation note, which also stops the agent.

### Does turning Autopilot off delete anything?

No. It only stops future automatic replies in that conversation. Nothing already sent changes, and the customer sees nothing.

### Should I run with Autopilot on from day one?

Test first. Leave the global **Active** switch off and use the AI Center's Live Test until you are happy with the answers — see [Test and activate your agent](https://support.unifunl.io/en/help/ai-center/test-and-activate).

## What's next?

Learn what happens when the agent decides a human is needed.

[Escalation](https://support.unifunl.io/en/help/inbox/escalation)

## Related articles

- [Using the inbox](https://support.unifunl.io/en/help/inbox/using-the-inbox.md): Work every conversation from one screen — the list on the left, the exchange in the middle, and the contact record on the right.
- [Escalation](https://support.unifunl.io/en/help/inbox/escalation.md): What happens when the agent hands a conversation to a human — the yellow note, the silent agent, and how to pick it up.
- [Test and activate your agent](https://support.unifunl.io/en/help/ai-center/test-and-activate.md): Rehearse with Live Test, then go live with the Active switch — and understand how it differs from AI Autopilot.