Part 2 Reliable agents
Getting the right answer is step 1. The real power comes when your AI can act on it: book a flight, update a database, resolve a support ticket, and coordinate with other models to handle an entire workflow end to end. That’s when things get exciting and dangerous. A wrong answer is an inconvenience, but a wrong action is a production incident.
This part of the book moves from outputs to actions. The goal is agents that not only function but also genuinely help people get things done: book the right flight, find the right product, and resolve the right ticket without breaking things along the way.
Chapter 6 introduces the architecture of reliable agents: memory systems, tool integration, the ReAct reasoning framework, and guardrails that keep agents on track. Chapter 7 covers tool integration and Model Context Protocol (MCP), including the emerging agent harness pattern, which separates the control plane from the compute plane for security and crash recovery. Chapter 8 brings everything together with multi-agent systems; you’ll build ShopBot, a LangGraph-powered e-commerce assistant with specialized agents for intent classification, product search, and question-answering.
By the end of part 2, you’ll know how to build agents that real users rely on: systems that handle the messy, unpredictable requests people throw at them every day.