The agent and its limits
In short. The agent works inside limits the system enforces, hands sensitive topics to a person every time, and can be paused at any moment.
What the agent does
The agent reads the ticket, the customer’s account and the procedures your team has written for it. It looks things up and takes small actions, and every step it takes is recorded.
What it never does alone
These topics always go to a person, with a summary and a draft reply:
- refunds and credits
- legal threats
- complaints about staff
- security incidents
- account deletion
- anything involving minors
Limits the system enforces
Every action the agent can take has its own limits, for example once per ticket. The system checks those limits itself. Even if a customer tries to talk the agent into more, the action is refused.
Each piece of work also has a limit on steps, cost and time. If the agent reaches one, it hands the ticket to a person. It never just stops.
Visitors who aren’t signed in
Someone the desk can’t confirm as the account holder only gets answers from public information. For anything about their account, they’re asked to sign in or to reply from the account’s email address.
How you stay in control
- Shadow: the agent drafts replies on real tickets and sends nothing. Your team reviews the drafts.
- Live: the agent replies to customers.
- Paused: the agent doesn’t pick up new tickets, and work it hasn’t finished is cancelled.
You choose the mode for each brand separately on the Agents screen. A person can also take over any ticket at any time. If the agent was about to work on it, that work is cancelled.
Seeing why it did something
Open any piece of the agent’s work to see each step: what it read, what it did, and why it stopped.
Customers can always tell when a reply came from the AI, and they can always ask for a person.