The AI agent in support
The agent is the AI component that takes part in a case. There are two ways to put it to work, and the difference between them decides whether the customer on the other side talks to the AI at all.
Stage agent and supervisor agent
A stage agent is assigned to a stage of the funnel. It acts on cases sitting in that stage and can reply to the contact.
A supervisor agent is bound to the case, not to the stage. It does not talk to the contact: it talks to your team, inside the case, through Chat com IA. It is there to look up history, summarise the conversation and suggest the next action — always with a person deciding what goes out.
An agent configured as a supervisor takes no stage assignment. That is deliberate: the two roles are mutually exclusive, and allowing both at once would produce an agent that replies to the customer while advertising itself as an internal advisor.
Chat com IA
Chat com IA lives inside the case. It can see that case's conversation and the contact's data, and it answers your team.
What comes out of it is not sent to the contact. To send something, a person copies it, reviews it and sends — the manual step is the control.
Playbook: what the agent is allowed to do
The agent's behaviour comes from a playbook — the decision tree it follows. The playbook is versioned: publishing a new version does not erase the previous one, and you can roll back.
The playbook is always previewed before publishing. The preview renders the flow as a diagram, and it is where you notice an orphan stage or a question that is never reached.
Limits worth knowing
- The agent answers from what is in the case. Information that only exists in the team's heads, or in a system that is not integrated, it does not have.
- A supervisor agent does not replace reading the case when the conversation is long: it summarises, and a summary loses detail.
- Every generated text passes through human review before reaching the contact. There is no mode in which the supervisor sends on its own.