Agent modes and human handoff
The four conversation modes and how the agent escalates to a human.
Every WhatsApp conversation runs in one of four modes. The mode decides who is replying right now: the AI on its own, the AI with your review, you, or nobody. You can change it at any time, and the AI can hand a chat back to you when a customer needs a person.
You set the mode from the badge at the top of the conversation. It reads the current state and lets you switch with one click.
The four modes
| Field | Type | Description |
|---|---|---|
| AutonomousAI sends | AI handling | The AI replies to the customer automatically, with no review. This is hands-off mode. |
| AssistedYou approve | AI drafting | The AI writes a suggested reply, but nothing is sent until a human approves or edits it. |
| ManualYou reply | You're in control | The AI is paused. Only your team replies in this thread. |
| ResolvedClosed | Resolved | The conversation is closed. Nothing replies. Reopening happens automatically when the customer messages again. |
Switching is instant and reversible
Change modes as often as you like. The badge color and label update the moment you switch, and the change is shared with your whole team in real time.
When you switch a conversation to Resolved, MAZZA fires the whatsapp_resolved automation trigger. You can use that to send a follow-up, log the outcome, or tidy up the record. See Automations overview.
When the AI hands a chat to a human
The AI does not try to fake its way through situations it should not handle. When a customer asks for a person, or the AI hits one of its built-in guardrails, it escalates: it flags the conversation, sends one short handoff line to the customer, and pulls your team in.
The AI escalates when:
- The customer explicitly asks for a human, manager, owner, or a callback. Your team is paged urgently.
- The customer is angry, complaining, or threatening action.
- The question is outside the AI's knowledge or tools, such as legal matters, accidents, fines, or refunds.
- The AI fails to understand the customer twice in a row.
- The customer mentions any extra escalation keyword you set in Agent Studio.
The flagged conversation appears under the Needs you filter in the inbox, alongside any manual or assisted chats with unread messages. A banner inside the thread tells you exactly why the AI handed it over, in plain language.
On failure, the agent holds the line. It never sends a silent draft.
If the AI ever cannot complete a reply, MAZZA does one thing, never the other. It sends the customer a short holding message so they are not left on read, hands the chat to a human in Manual mode, and pages your team. It does not quietly save an unsent draft for you to find later. Drafts are an opt-in feature of Assisted mode, never a failure fallback. See Assisted mode: reviewing AI drafts.
There is one important exception that protects the customer experience. When a customer explicitly asks for a human, the handoff confirmation is sent immediately, even in Assisted mode. A customer waiting for a person should never sit behind a draft review.
Clearing the flag
A flagged conversation stays flagged until you re-enable the AI. Switching the mode back to Autonomous or Assisted clears the escalation flag, and the chat drops off the Needs you filter at once.
Open the flagged chat
Find it under the Needs you filter, or by the alert on the conversation.
Read the reason and reply
The thread banner explains why the AI handed it over. Reply as yourself in Manual mode.
Hand it back when you are done
Switch the mode to Autonomous or Assisted. This clears the flag and lets the AI resume.
Set a default, then adjust per chat
Most teams pick one default mode for new conversations and only switch individual chats when needed. You can change the default and your escalation rules in Agent Studio. To learn the difference between drafts and full autonomy, read Assisted mode: reviewing AI drafts.
Common questions
What happens to a resolved chat if the customer replies?
It reopens automatically. The customer's new message lands in your inbox like any other, and the conversation picks up its mode again.
Does Manual mode turn the AI off for good?
No. Manual mode only pauses the AI for that one conversation. Switch back to Autonomous or Assisted and the AI resumes there. Other chats are unaffected.
Can the AI escalate while in Assisted mode?
Yes. The AI still flags conversations and pages your team. If the customer directly asked for a human, the holding line is sent right away rather than held as a draft.
Where do escalation alerts go?
To your configured team channels and the Needs you filter. Set up where you get pinged in Notifications and team alerts.
Last updated June 2026