A practical guide to debugging AI agents in production: how to isolate failures across prompts, tools, memory, models, and runtime state without wasting days on vague theory.