When the agent hands over

The agent is built to stop rather than guess. A handover is a designed outcome, not a failure — and reading them as failures is the most common mistake new teams make with it.

What triggers a handover

The agent is unsure. Every answer carries a confidence, and below your threshold it asks for a person instead of sending.

A safety check rejected the draft. Two run before anything is sent: one verifies that every product and every id in the answer actually came from a real lookup, the other checks for disclosed personal data, an overridden price, a prohibited topic, wrong language or excessive length. A draft that fails either is not sent.

The customer asks for a human. Directly, or by being unhappy enough that continuing would make it worse.

A request needs authority the agent does not have. A discount, an exception to your terms, anything about money that is not already decided.

You built a rule that says so. Any workflow can hand over — see Triggers and actions.

What you get

The whole conversation, not a ticket. Whoever picks it up sees what the customer asked, what the agent answered, what it looked up, and why it stopped. The customer is not asked to repeat themselves — which is the single biggest reason handovers feel bad when a tool does them badly.

Escalations can be assigned automatically: round-robin, or to whoever currently has the fewest open conversations.

The agent stops answering that conversation

When a person takes over, auto-replies pause for that conversation. Two voices answering one customer is worse than a slow answer, so this is not configurable per message — it follows the handover.

Replying from an assistant over MCP pauses it too, for the same reason.

Tuning it

Too many handovers usually means the agent lacks knowledge, not that the threshold is wrong. Before lowering confidence, check what it was unsure about — the answer is often a missing page in your knowledge base, and adding it removes a whole class of handovers at once.

Too few means the threshold is loose, and you will see it as customers correcting the agent rather than as escalations. That is the more expensive direction.

Start strict. An agent that hands over too often costs you time; one that answers confidently and wrongly costs you a customer.