Sometimes the AI isn't the right voice for a conversation. The user asked for a human. They're frustrated. The topic is sensitive. They're asking about something the agent doesn't know. When that happens, take the thread over manually.
Handing off
Click I'll take it from here in one of two places:
- On the Draft Card, next to Send reply.
- In the thread header, where it appears as a Hand off button.
A few things change at once:
- The current draft is discarded.
- The AI stops drafting on this conversation. Even if the user sends more messages, no new draft is generated.
- The conversation row in the list picks up the user-check icon and moves into the Handed off filter tab.
- The thread header shows a small Manual mode tag next to the user's name.
- The Draft Card is replaced by a Manual Compose box at the bottom of the thread.
The Manual Compose box
It's a plain textarea. Type your reply, press Send reply (or Cmd+Enter / Ctrl+Enter), and the message goes straight to Meta. There's a character counter on the right that turns amber when you're approaching the platform's limit:
- Messenger: 2000 characters.
- Instagram: 1000 characters.
If something goes wrong on Meta's side, the textarea restores what you typed so you can fix and retry without retyping.
The label below the textarea reads Manual mode · Cmd+Enter to send, which is also a reminder of where you are.
Giving the conversation back to the AI
If you took the conversation over but realize the AI would actually do fine, click Reopen for AI in the thread header. The Manual Compose box goes away. The next inbound message from the user will get an AI draft as usual.
This is one-way safe: handing off and reopening as many times as you want doesn't lose history.
The reply window meter
The thin colored bar in the thread header is the 24-hour reply window. Meta only lets you send a standard message within 24 hours of the user's last message. The meter shows how much of that window is left:
- Green — plenty of time.
- Amber — under half left.
- Red — about to close.
The label next to the bar reads "X h Y m left to reply." Once it crosses zero, the label changes to Reply window closed and Send disables.
The window resets every time the user sends a new message. If they re-engage, you're back in green.
There's no way to send a standard message after the window closes during the beta. Higher-tier message types (Meta calls these message tags or Human Agent permissions) are something we may add later.
When to hand off (a short list)
- The user explicitly asked for a person. Don't make them ask twice.
- You're about to share something sensitive (financial detail, account-specific data, legal language).
- The AI looks like it's missing context the user clearly expects you to have.
- The thread is heading toward an apology or a refund. AI apologies tend to read as hollow.
- The conversation is high-stakes (an influencer, a journalist, a complaining customer with a large audience).
For everything else, the AI draft is a fine starting point. Edit and send.
Next steps
NEXT ARTICLE