Design a support AI agent with policies, escalations, and evaluation checks. [Editorial sample]
Design a customer support AI agent for human-in-the-loop use.
Product:
{{product}}
Support channels:
{{channels}}
Top ticket types:
{{ticket_types}}
Policies / red lines:
{{policies}}
Create:
1) Agent purpose and non-goals.
2) Intent taxonomy for top ticket types.
3) Allowed tools/actions vs forbidden actions.
4) Escalation rules (when to hand off).
5) Draft system prompt skeleton.
6) Evaluation checklist (accuracy, tone, policy adherence) — no fake benchmark scores.
7) Rollout plan: shadow → suggest → limited auto for FAQs only.
Emphasize that refunds, legal, and security issues stay human-owned.Run with your support lead. Use output as a design doc, not a production system prompt dump.
B2B SaaS; Intercom + email; password resets, billing questions, bug reports.
Design doc with intents, escalations, and rollout stages. (Illustrative.)