AI Agents in Action, Second Edition
by Manning (Micheal Lanham)
Build, evaluate and deploy Python agents on the OpenAI Agents SDK and MCP, from a first tool-using agent to multi-agent research loops.
Overview
AI Agents in Action, Second Edition by Micheal Lanham was published by Manning in June 2026 as a 392-page book, and Manning describes it as a near-total rewrite of the 2025 first edition, which was built on GPT Assistants and early AutoGen. The new edition frames an agent as five functional layers (persona, actions, reasoning, memory and evaluation) and runs in eleven chapters. Chapters 1 and 2 cover agentic thinking, LLM and prompting basics and a first agent on the OpenAI Agents SDK. Chapter 3 is on actions through the Model Context Protocol, including building your own MCP servers. Chapter 4 covers multi-agent architecture, handoffs and guardrails. Chapter 5 covers reasoning and planning patterns such as ReAct, Reflexion, Tree-of-Thought and the Sequential Thinking MCP server. Chapter 6 adds vector search, RAG knowledge agents and MCP-backed memory. Chapter 7 covers test-driven agent development, critic and grounding agents and Phoenix tracing. Chapter 8 covers Docker deployment, security and governance. Chapters 9 to 11 cover the agentic loop with a deep research agent, a metacognitive cognitive agent and field-tested tips for support, RAG and research agents. The companion repository (Apache-2.0, Python 3.11 or later, Node.js 20 for npx-launched MCP servers) holds code for chapters 2 to 11 and pins the MCP Python SDK below 2.0 because the chapter 3 and 4 servers use FastMCP. Lanham has more than 20 years in software and also wrote Manning's Evolutionary Deep Learning.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Book
- Cost
- Paid
- Duration
- 392-page book, 11 chapters with end-of-chapter exercises, self-paced
- Provider
- Manning (Micheal Lanham)
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Build a first agent with the OpenAI Agents SDK and extend it with function tools
- ✓Use existing MCP servers from an agent and write your own MCP servers with FastMCP
- ✓Architect multi-agent systems with agent flows, handoffs and guardrails that retry on failure
- ✓Apply ReAct, Reflexion, Tree-of-Thought and the Sequential Thinking MCP server to agent planning
- ✓Build RAG knowledge agents on a vector database and add persistent memory through MCP
- ✓Evaluate agents with test-driven development, critic and grounding agents and Phoenix tracing
- ✓Containerize agent systems with Docker Compose and plan security, safety and governance for production
- ✓Implement a deep research agent and multi-agent orchestration loops that search the web through MCP
Highlights
- •Second edition rewritten around the OpenAI Agents SDK and MCP, replacing the first edition's GPT Assistants and AutoGen code
- •Evaluation gets a whole chapter (test-driven agent development plus Phoenix tracing) instead of a closing aside
- •Apache-2.0 companion repo with per-chapter code, an ERRATA file and a check_repo.py script that pre-downloads the MCP servers the examples use
- •Rated 4.6 from 44 Amazon US ratings as of October 3, 2026, per a Computing for Geeks roundup that names it the pick for broad agent coverage
Who It’s For
Best For
- ✓Python developers building their first production agent on the OpenAI Agents SDK
- ✓Engineers who need to write and consume MCP servers rather than only call hosted tools
- ✓Teams adding evaluation, tracing and deployment discipline to agent prototypes
- ✓Readers of the first edition whose AutoGen and Assistants code no longer matches current SDKs
Prerequisites
- •Intermediate Python (Manning's stated audience); no prior agent experience required
- •An OpenAI API key for all examples and a Brave Search API key for the chapter 9 and 10 research agents
- •Python 3.11 or later, Node.js 20 or later for npx-launched MCP servers, and Docker for the Phoenix and deployment chapters
FAQ
What is AI Agents in Action, Second Edition?
AI Agents in Action, Second Edition is a 2026 Manning book for intermediate Python developers who want to build AI agents end to end. It starts with a single agent on the OpenAI Agents SDK and works up to MCP tool servers, multi-agent handoffs, RAG memory, Phoenix-based evaluation and Dockerized deployment. You finish able to ship a deep research agent and a multi-agent workflow you can test.
Is AI Agents in Action, Second Edition free?
AI Agents in Action, Second Edition is a paid resource.
What level is AI Agents in Action, Second Edition for?
AI Agents in Action, Second Edition is aimed at a intermediate audience. Recommended background: Intermediate Python (Manning's stated audience); no prior agent experience required, An OpenAI API key for all examples and a Brave Search API key for the chapter 9 and 10 research agents, Python 3.11 or later, Node.js 20 or later for npx-launched MCP servers, and Docker for the Phoenix and deployment chapters.
How long does AI Agents in Action, Second Edition take?
Expect roughly 392-page book, 11 chapters with end-of-chapter exercises, self-paced. Most learners work through it at their own pace.
What will I learn from AI Agents in Action, Second Edition?
You'll learn: Build a first agent with the OpenAI Agents SDK and extend it with function tools; Use existing MCP servers from an agent and write your own MCP servers with FastMCP; Architect multi-agent systems with agent flows, handoffs and guardrails that retry on failure; Apply ReAct, Reflexion, Tree-of-Thought and the Sequential Thinking MCP server to agent planning; Build RAG knowledge agents on a vector database and add persistent memory through MCP; Evaluate agents with test-driven development, critic and grounding agents and Phoenix tracing; Containerize agent systems with Docker Compose and plan security, safety and governance for production; Implement a deep research agent and multi-agent orchestration loops that search the web through MCP.
Topics
Sources
This page was written from 4 sources, 2 on domains other than manning.com.