The Agentic Engineer: Designing, Evaluating, and Operating Multi-Agent AI Systems in Production (Paperback) (ISBN-13: 9798184943411)

Most AI agents work great in a demo and fall apart in production. This book is about closing that gap.

The Agentic Engineer is a build-along guide for software engineers, architects, and technical leads who have to design, ship, and own agentic systems - not just talk about them. If you can write code and you've made a few LLM calls, you have everything you need to start. You don't need a machine-learning background, and you won't find keynote hype here. You'll find the engineering: the harness, the loop, the guardrails, the evals, and the hard tradeoffs most tutorials skip.

The book is anchored to one running project. You start with a single graph node and one tool call, and across thirteen chapters you build it into a complete, production-shaped agent - one that can search and reason, remember across sessions, delegate to subagents, enforce guardrails at every layer, pause for human approval on risky actions, and emit a full trace of everything it did and why. The stack is LangGraph, Langfuse, and Python, chosen because they expose the seams that matter instead of hiding them.

By the end, you'll be able to:

• Explain what actually separates an agent from a prompt chain - and recognize when you don't need an agent at all
• Engineer the harness and control loop that make an agent terminate cleanly instead of looping forever or silently burning money
• Implement the core patterns - ReAct, Plan-and-Execute, Reflection - and know which one fits which problem
• Design tools an agent can use reliably, and manage memory and state without drowning in context
• Add input, output, and tool-call guardrails, and orchestrate multiple agents without chaos
• Instrument full observability, build evaluations that catch regressions, and harden for production with retries, fallbacks, and cost controls
• Reason about safety, trust, and prompt injection at the system level

Every chapter ends with the failure modes engineers actually hit - the anti-patterns - because what breaks teaches faster than what works. The final chapter walks two complete case studies, a customer support agent and a coding agent, including what went wrong in production and why.

A companion GitHub repository tracks the book chapter by chapter, so you can build alongside the text.

If you've been asked to "add an AI agent" and you suspect the honest answer is more complicated than the slide deck implied - this is the book for you.