All playbooks / System Design

Playbook · System Design

Design an AI-powered customer support chatbot.

The interviewer is not looking for a vendor shopping list. They want to see whether you understand that support bots sit on top of trust, policy, and customer operations. The strong answer separates informational help from account-changing actions and explains how the system knows when to stop pretending and escalate.

Senior High frequency 11 min read Premium
Practical answer framework for AI engineer interview loops.

01Interview Context

The interviewer is not looking for a vendor shopping list. They want to see whether you understand that support bots sit on top of trust, policy, and customer operations. The strong answer separates informational help from account-changing actions and explains how the system knows when to stop pretending and escalate.

02The 90-second answer

I would design the system as layered paths: channel ingress, orchestration, retrieval for informational answers, action tools for authenticated account work, policy checks, and human escalation. The central design choice is that read-only help and write actions should not share the same safety posture. Informational answers can move quickly with confidence gates. Account-changing actions need identity, permissions, audit logs, and rollback-safe APIs.

Next playbook

What is an AI agent, and how does it differ from a simple LLM call?

14 min · AI Agents