Summary

Michael Albada, a machine learning engineer who has built large-scale multi-agent systems at Uber, ServiceNow, and Microsoft, structures this as a full lifecycle for agent applications rather than a tour of any single framework. It covers the core components of an agent, tools, memory, and orchestration, works through popular frameworks including LangGraph, AutoGen, and CrewAI, then moves into coordination patterns for multi-agent systems, evaluation, production monitoring, security, and human-agent collaboration. The throughline is that going from a single agent to a coordinated multi-agent system introduces failure modes, evaluation problems, and security concerns that don't exist in a single-agent prototype, and the book treats each as a design decision rather than an afterthought.

Target Readers

• Backend engineers who have shipped a single LLM agent and are now designing a multi-agent system that has coordination problems the prototype didn't have
• Architects evaluating agent frameworks like LangGraph, AutoGen, or CrewAI who want the tradeoffs, not a single recommended stack
• Engineers responsible for production monitoring, evaluation, or security of an agent system who need those treated as first-class design concerns

Colophon

Publisher

オライリー・ジャパン

ISBN

978-4-8144-0159-8

Published

Apr 2026

List price

¥4,620incl. taxMay differ from the actual selling price on Amazon