Architecting Multi-Agent Systems (Paperback)

Paperback. The agentic AI space is accelerating fast - and most engineers are shipping fragile prototypes they will be rewriting in six months. They are locked into single frameworks, missing the foundational reasoning behind orchestration decisions, and finding that agents that demo beautifully collapse under real-world load.Architecting Multi-Agent Systems closes that gap. Taking a first-principles approach, this book builds the PicoAgents framework from scratch so you understand exactly what AutoGen, LangGraph, and CrewAI are doing under the hood - then teaches you to design systems that are production-ready, enterprise-scalable, and framework-agnostic by design. What you will learn: Build multi-agent systems from first principles using the PicoAgents library, so no framework is a black box to you againImplement Model Context Protocol (MCP) and Agent-to-Agent (A2A) protocols for interoperable, extensible agent communicationDesign production-grade RAG pipelines and agent memory architectures that survive real workloadsApply orchestration patterns - routing, reflection, tool use, and handoffs - across complex multi-agent workflowsEvaluate and test your agents systematically with structured benchmarks before you ship, not afterDeploy, observe, and scale agentic AI in enterprise environments using LLMOps and observability best practicesNavigate governance, cost optimisation, and compliance for enterprise-grade agentic AI deploymentsWritten by a Microsoft Research contributor and AutoGen core team member, this is the book for software engineers, ML engineers, and technical leads who have moved past what an AI agent is and need to ship systems that hold up in the real world. Whether you are architecting your first production This item is printed on demand. Shipping may be from multiple locations in the US or from the UK, depending on stock availability.…