Multi-Agent AI Engineering: Design, build, and operate AI systems that think and act as coordinated teams

Move agentic AI from clever demos to reliable production with the principles, patterns, and practices behind multi-agent systems at scale.

Key Features

• Build production-ready agents and multi-agent systems with hands-on Python examples
• Apply foundational principles, proven design patterns, and orchestration strategies
• Evaluate, observe, scale, and evolve agent systems through real-world case studies
• Purchase of the print or Kindle book includes a free PDF eBook

Book Description

As AI systems take on more complex tasks, the limits of single-model applications become increasingly clear. Problems requiring long-horizon reasoning, specialized expertise, coordination, and parallel execution demand multiple agents working together reliably in production.

But building multi-agent systems is fundamentally an engineering challenge. Agents must communicate, delegate tasks, manage context, recover from failures, and stay aligned on shared goals under real-world constraints.

Multi-Agent AI Engineering is a practical guide to designing and operating production-grade multi-agent systems. Drawing on the authors’ research, open-source contributions, and experience building AI systems at scale, the book focuses on architectural principles that extend beyond any single framework or trend.

You’ll explore agent foundations, communication protocols, memory and context management, orchestration, interoperability standards, and canonical multi-agent patterns through hands-on Python examples. The book also covers production realities including evaluation, observability, reliability, safe self-improvement, and scaling agentic systems in practice.

By the end, you’ll be equipped to design, build, and scale reliable multi-agent systems for real-world deployment.

What you will learn

• Apply foundational principles to design production-ready agents
• Design agent communication, routing, and collaboration flows
• Orchestrate teams with proven multi-agent design patterns
• Manage memory, retrieval, and context across agent teams
• Evaluate, red-team, and benchmark agent system behaviors
• Deploy and scale multi-agent systems in production
• Instrument agents with OpenTelemetry-based observability
• Evolve and improve agent systems safely in production

Who this book is for

If you are an AI engineer, ML practitioner, software architect, or technical leader who wants to move beyond agent demos and ship multi-agent AI systems that work in production, this book is for you. By the end, you will be able to design, deploy, evaluate, and continuously improve agentic systems with confidence. It is equally valuable for engineering and product managers making informed decisions about agentic AI architecture. Readers should be comfortable with Python and have basic familiarity with LLMs; deep ML expertise is not required.

Table of Contents

1. Introduction to Multi-Agent Systems
2. Principles of Multi-Agent Systems
3. Frameworks and Mental Models
4. Constructing Your First Agents
5. Agent Communication and Collaboration
6. Context Management in Agents
7. Orchestrating Agent Teams
8. Unified Abstractions and Protocols for Agent Collaboration
9. Design Patterns for Multi-Agent Collaboration
10. Comparative Survey of Frameworks
11.  Evaluating the Performance and Behaviors of Multi-Agent Systems
12. Deploying and Scaling Multi-Agent Systems
13. Observability for Agentic AI
14. Case Studies from Real-World Applications
15. Conclusions and Future Outlook

Xiao Ma is an engineering executive with 15+ years of experience in ML/AI systems across academia and industry. He holds a Ph.D. in Computer Science from the University of Illinois, specialized in the intersection of machine learning and computer systems. As Chief Architect at Pattern Insight and Medium, he led teams developing large-scale ML systems for both enterprise and consumer markets. Currently at Splunk, a Cisco Company, he leads teams building Splunk Observability Cloud, a full-stack solution (connecting infrastructure, applications, and business impact) featuring enterprise-class multi-agent AI systems and industry-leading AI Observability products that empower customers to innovate with confidence at scale.

Chi Wang is the creator of AutoGen, AG2, MassGen, Sutando — open-source projects for agentic AI used by Nvidia, Google, Microsoft, and leading research institutions worldwide. He previously led agentic AI work as a Senior Staff Research Scientist at Google DeepMind, pioneered agentic AI research at Microsoft Research, and created FLAML for AutoML. He teaches at Stanford, Berkeley, Coursera, and DeepLearning.AI. His work has earned the UIUC Siebel School Early Career Alumni Achievement Award, Best Paper at the ICLR'24 LLM Agents Workshop, and the SIGKDD PhD Dissertation Award.