AI Employee roles and autonomy levels
AI Employees
Roles
| Role | Handles | | --- | --- | | sales | purchase interest, pricing questions, appointment requests | | support |
support requests, complaints, refunds | | appointment | books appointments when a real slot is given | |
lead_qualification | qualifies purchase/pricing interest | | review | triages complaints and support |
Autonomy levels
| Level | Behaviour | | --- | --- | | observe | Classifies only; records a decision, never acts | | suggest | Records a
decision and policy verdict; never acts | | assist | Allowed actions wait in AI Approvals for a human | | act | Executes
permitted tools automatically | | autonomous | Executes with the widest policy envelope |
Customer-facing replies
Switch an employee to customer-facing to let it answer allowlisted intents immediately (general information, pricing,
product questions and appointment requests). Replies are fixed, approved text — the bot never improvises. Anything
outside the allowlist escalates to a human.
Tools
Tools are default-deny per employee: tick only what the role needs. Requesting a human review is always the safe
fallback.