You already know: the system design method, enterprise RAG and the agent platform ideas from the LLM platforms case. This case adds the thing that makes agents genuinely hard: actions with consequences. An answer can be wrong; a refund, once issued, is money gone.
How to read this lesson. No code, on purpose. Numbers marked assumed are assumptions you state out loud in the interview; prices are illustrative.
1. The prompt, and the questions that change the design
Most candidates hear "chatbot with tools". A staff candidate hears four harder problems: who is allowed to do what (the customer, the agent, the policy), how to keep a long conversation's state correct, when to give up and hand off, and how to measure success without being fooled by deflection.
Assumptions you state: