Agentic AI Engineering

You understand how agents work under the hood. Now build them the way professional engineering teams actually do.

Anyone can copy a LangChain tutorial. Far fewer people understand what that framework is actually doing for them — which means far fewer people can debug it when it breaks, deploy it responsibly, or know when a simpler approach would serve them better.

AGENTIC AI ENGINEERING is a 16-chapter, hands-on guide to the real toolchain behind production AI agents: LangChain, the Model Context Protocol (MCP), LangGraph, and FastAPI. Every concept is taught by first showing you the mechanism underneath it — so the framework never feels like unexplained magic.

You'll build Atlas, a complete research agent, growing it chapter by chapter from a single LCEL chain into a fully deployed, tool-using, memory-persistent web service.

What you'll actually build:

• Composable pipelines with LangChain Expression Language (LCEL) — and exactly what the pipe operator is doing


• Persistent, multi-session memory with automatic history management


• Tool-calling agents — built manually first, then automated with AgentExecutor


• Reliable structured output using Pydantic, instead of hoping a model formats JSON correctly


• Real production resilience: retries, exponential backoff, circuit breakers, and graceful fallbacks


• A working MCP server and client — the emerging standard for connecting agents to tools


• Multi-step orchestration with LangGraph, including cycles, conditional routing, and human-in-the-loop checkpoints


• Full observability with LangSmith tracing, plus a rigorous, automated evaluation suite


• A real deployment: Atlas as a FastAPI web service, with authentication, streaming, and proper error handling

Two complete capstone projects close out the book: a fully guided build (TripPlanner, an MCP-powered trip-budgeting agent) and a lighter-scaffolded challenge (PracticeTrack, a habit-tracking agent) where you apply the full toolchain to a new problem on your own.

Who this book is for:
Developers who've already built an agent from raw API calls — whether from Book One in this series or equivalent hands-on experience — and are ready to learn the professional framework ecosystem without losing the understanding that makes frameworks genuinely useful rather than a black box.

Written by Dave Malhotra, with the same commitment to real, working code over slideware: every chapter builds something you can actually run, tests your understanding, and connects directly to what came before it.

A free companion package — full source code for Atlas, TripPlanner, PracticeTrack, and their MCP servers — is available for download, no registration required.

This is Book Two in the Agentic AI Engineering series. Book Three goes further, into multi-agent orchestration, retrieval-augmented generation, and long-term memory — unifying everything into one complete, intelligent system.

Learn the tools the way engineers who actually understand them do.