Building Agentic AI Systems

Go beyond standalone agents. Build interoperable, production-ready agent systems with MCP, A2A, agent loops, harness engineering, evaluation, observability, safety, and enterprise cloud deployment.

Key Features

• New in the second edition: connect agents with MCP, A2A, and modern multi-agent frameworks
• Engineer reliable agent loops and harnesses for memory, tools, context, retries, and guardrails
• Evaluate, secure, observe, and deploy agents across Kubernetes and managed AI platforms

Book Description

The first edition of Building Agentic AI Systems showed how autonomous agents reason, plan, reflect, use tools, and collaborate. This second edition tackles the next challenge: connecting agents across tools, frameworks, and platforms, then engineering the loops, harnesses, and infrastructure for reliable production. You’ll strengthen the agent stack with memory, context engineering, reasoning and recovery loops, and harness engineering, then move into the protocols reshaping agentic AI. Implement Model Context Protocol (MCP) for tool and data access, use A2A for capability discovery and cross-agent collaboration, and understand where it fits for asynchronous agent communication. New coverage brings these standards together with LangGraph, CrewAI, Google ADK, and Strands. You’ll design multi-agent workflows and deploy distributed systems with Kubernetes, Amazon Bedrock AgentCore, and Google Vertex AI Agent Engine while addressing tracing, debugging, cost, reliability, and trust boundaries. You’ll also evaluate agents with AgentBench, GAIA, SWE-bench, and production metrics, apply human-in-the-loop controls and governance, and assess where Small Language Models can reduce latency and cost. By the end, you’ll build connected agent systems that are interoperable, observable, secure, measurable, and ready to scale.

What you will learn

• Design interoperable multi-agent systems with MCP and A2A
• Engineer agent loops for reasoning, reflection, tools, and recovery
• Build harnesses for context, memory, tool access, and guardrails
• Integrate LangGraph, CrewAI, ADK, Strands, and agent frameworks
• Deploy distributed agents on Kubernetes and managed AI platforms
• Evaluate agents with AgentBench, GAIA, SWE-bench, and tracing
• Apply HITL, governance, safety controls, and trust boundaries
• Use SLMs for lower latency, edge deployment, and cost control

Who this book is for

This book is for software developers, AI engineers, enterprise and solution architects, platform engineers, and technical leaders building production agentic systems. It is ideal for readers of the first edition who want to move from standalone agents to interoperable multi-agent architectures, agent loops, harness engineering, modern protocols, evaluation, and cloud deployment. Familiarity with Python, APIs, LLM applications, or basic agent concepts is helpful.