Train your knowledge

Building Procedures the agent can run safely

In short: Define which procedures the agent may run, set permission boundaries for every action, and require human approval before any irreversible action.

Procedures let the agent act within limits you set instead of deciding for itself which actions are allowed. A safe setup starts by choosing the actions the agent should perform, then making sure each action stays within its approved boundary.

Choose the actions to allow

Procedures can cover specific actions, such as pilot signups, demo bookings, or custom-quote handoffs. The Customer Agent can also run approved workflows, update profiles, file requests, change subscriptions, and route tickets. Only include actions you have approved for the agent to perform.

Set permission boundaries

Every action should have a permission boundary, so the agent runs only procedures you have defined. No irreversible action should happen without a human approving the boundary first. The Customer Agent’s procedures are described as permission-based and reversible.

Keep an audit trail

An audit trail records procedure runs and agent actions. For the Acquisition agent, the trail captures every run for compliance and review; the Customer Agent logs actions with a full audit trail. Use these records to review what the agent did and confirm it stayed within its approved scope.

The available information describes procedure examples and safeguards, but does not provide step-by-step instructions for creating procedures in the product. Defining the permitted actions, setting boundaries, and retaining an audit trail are the safety principles it does specify.