How Do You Evaluate an Agent That Calls Tools That Call Other Tools? | HackerNoon

A good final answer can hide failures elsewhere in an AI agent. Test tool routing, evidence retrieval, rules, and model output separately