[Book Review] Amazon Bedrock AgentCore Practical Introduction ─ A Book for Building and Operating AI Agents on AWS

Introduction

Hello, I'm Jinno from the consulting division, and I'm a huge fan of Amazon Bedrock AgentCore.

This time, I'd like to introduce the book Amazon Bedrock AgentCore: A Practical Introduction, which will be released on May 29, 2026!

https://www.sbcr.jp/product/4815641238/

I was fortunate enough to be invited as a technical reviewer for this book, which gave me the opportunity to read the manuscript from the writing stage. Throughout the review process, I repeatedly felt how amazing this book was, with so much of my beloved AgentCore knowledge packed into it.

This is my first time writing a book review blog, but as someone who read through all the chapters, I hope to share my honest impressions!

Book Overview

All three authors are also the authors of the bestselling previous book, Amazon Bedrock: Introduction to Generative AI App Development.

They are people who are active on the front lines in the fields of generative AI and AI agents, and I always look up to them. When I first read Amazon Bedrock 生成AIアプリ開発入門, I knew nothing about Bedrock, so being able to catch up through that book in an easy-to-understand way made me unilaterally admire all three of them. I never even dreamed that I would one day be reviewing the follow-up book by these same people, and I am truly grateful that such a day has come.

Target Readers

The book's cover lists the following as "target readers":

• Those who want to learn how to build efficiently on AWS
• Those who want to take on the challenge of developing high-performance AI agents
• Those who want to learn through hands-on experience
• Those who want to learn systematically from the overall structure

While IT engineers and developers are the primary audience, the structure also considers those who have become interested in in-house development through AI. There are many sample code examples, and with AI tools like ChatGPT offering web search capabilities, even those who don't normally develop can steadily build their understanding by leveraging AI along the way. Actually building things is what leads to the most learning.

A sample code repository is also provided, so feel free to use it as a reference when building things yourself!

https://github.com/minorun365/agentcore-book

Table of Contents

The book is structured into 4 parts, 16 chapters, plus appendices.

Part 1: Fundamentals

• Chapter 1: Generative AI Basics and Introduction to Amazon Bedrock
• Chapter 2: Introduction to AI Agents

Part 2: Strands Agents

• Chapter 3: Introduction to Strands Agents
• Chapter 4: [Hands-On] Let's Build a Research Agent

Part 3: AgentCore

• Chapter 5: AgentCore Overview and Core Feature: "Runtime"
• Chapter 6: "Memory" for Managing Memories
• Chapter 7: "Identity" for Controlling External Authentication
• Chapter 8: "Gateway" for Bundling Tools
• Chapter 9: "Policy" for Controlling Tool Usage
• Chapter 10: "Built-in Tools" Unique to the Cloud
• Chapter 11: "Observability" for Visualizing Operational Status
• Chapter 12: "Evaluation" for Automated Quality Assessment
• Chapter 13: [Hands-On] Let's Build a Full-Stack Agent

Part 4: Advanced

• Chapter 14: Leveraging Internal Data for Agents with RAG
• Chapter 15: [Hands-On] Let's Build an Ambient Agent with CDK
• Chapter 16: Introducing AI Agents into Business Operations

Appendices

• Appendix 1: Setting Up an AWS Account
• Appendix 2: Setting Up a Development Environment
• Appendix 3: Bedrock Service Quotas

The AgentCore section in Part 3 alone accounts for approximately 40% of the entire book. True to its title of AgentCore: A Practical Introduction, one of the book's distinguishing features is that each AgentCore function is carefully covered in its own dedicated chapter.

My Personal Highlights!

Full Color with Abundant Diagrams — Exceptionally Easy to Understand

The first thing you notice when reading the book is the sheer number and quality of the diagrams.

AgentCore covers a wide range of features, and organizing each feature individually as well as understanding how they relate to each other can be quite challenging...

The book carefully explains these concepts with original diagrams, allowing you to intuitively grasp how each feature works and how they interact with one another.

Parts that are difficult to understand from official documentation alone become clear with the combination of diagrams and text — that "ah, so that's how it works" moment — which is extremely helpful.

Truly Keeping Up with the Very Latest Updates

There are many instances where you'll find yourself thinking, "they've even incorporated that update?" — I thought this was genuinely impressive.

While AgentCore and Strands Agents are rapidly evolving services, this book continuously incorporates the latest updates. New features such as AgentCore Payments (payment functionality from agents), AgentCore Optimization (continuous improvement based on evaluation results), HTTP target support for Gateway, and the OBO (On Behalf Of) flow for Identity are all reflected in columns and body text, and you can feel the dedication of the authors.

This field changes incredibly fast, so having everything systematically compiled as of May 2026 is itself extremely valuable, and no other book keeps up with developments as thoroughly as this one.

Hands-On Content That Dives All the Way Into SaaS Integration

This is my personal top recommendation!

In Chapter 13, you can experience a truly "full-stack" hands-on session where you build a calendar assistant with OAuth integration using the Google Calendar API, manage credentials with AgentCore Identity, and deploy a frontend with Next.js + Amplify.

When actually trying to introduce AI agents in a business setting, implementing a frontend and handling authentication integration with external SaaS products are unavoidable challenges. Hands-on content that tackles these issues head-on is rarely found in other books or online resources. The 3-Legged OAuth using Identity is particularly complex, but this hands-on section breaks it down in a very accessible way, allowing you to understand and build it yourself.

The frontend is also built with a modern UI, making it something you could actually incorporate into real business work, which is great.

Highlights by Section

Part 1: Fundamentals — Grasping the Big Picture of AI Agents

Chapter 1 starts from the basics of generative AI and Amazon Bedrock, and even includes hands-on content using the Converse API / ConverseStream API. It's structured so that even those new to Bedrock can read through it with confidence.

Chapter 2's introduction to AI agents covers a wide range of topics — from the ReAct pattern and agent design patterns to standard protocols like MCP and A2A that frequently come up. The diagrams are plentiful and easy to understand, so just this chapter alone gives you a pretty clear picture of "what are AI agents?"

Part 2: Strands Agents — Understanding the Framework

You can learn step by step, from the basic usage of Strands Agents to implementing multi-agent patterns.

The hands-on session in Chapter 4 involves building a research agent using the Swarm pattern (a pattern where multiple agents collaborate with each other). The mechanism by which five agents — PlanAgent, WhatsNewSearchAgent, RetrievalAgent, ReviewAgent, and OrchestratorAgent — work together is fascinating to watch as agents communicate and produce outputs in a near-futuristic way. Practical examples of Swarm like this are rare, making it very valuable as a reference. Differentiating models by using Opus, Sonnet, and Haiku according to each agent's role, and configuring prompt caching to reduce costs, struck me as very practical points.

Part 3: AgentCore — The Heart of the Book

This is the largest section of the book, explaining each of AgentCore's 8 features across dedicated chapters.

Each chapter follows a consistent structure: conceptual explanation → code examples → pricing. The fact that pricing is addressed in each chapter is a welcome feature, since it prevents the concern of "I'd like to try it, but what about costs?"

As the book itself notes, AgentCore is feature-rich, and you don't need to use all the features — an à la carte approach of adopting only what you need is the way to go. That's why it's important to understand each feature individually, and having them explained carefully one by one is a great help. Even features like Gateway, whose benefits aren't immediately obvious, are explained carefully — definitely worth checking out.

Part 4: Advanced — A Path to Business Implementation

Chapter 14 on RAG dives into practical territory with utilizing internal data using Bedrock Knowledge Base, while Chapter 15 covers building an ambient agent using CDK.

For document search, AI agents + RAG continue to be in demand, and the book carefully explains the configurations you'll want to know. It also touches on query decomposition and reranking methods for improving search accuracy in Knowledge Bases, introducing them in a way that lets you implement them as needed — which is very helpful. After all, building something is just the beginning — verifying and improving accuracy is also essential.

The "3-Step Recommended RAG Implementation Procedure" introduced in the column is also practical: first build simply with S3 Vectors, then gradually introduce Advanced RAG, then measure accuracy quantitatively with evaluation. Since it's easy to get lost on where to start with RAG, it's important to first try things out quickly and cheaply.

The ambient agent hands-on in Chapter 15 is also a highlight. Just by uploading a receipt image to S3, the agent automatically handles everything from image analysis to expense classification, sending approval request emails, and even creating Confluence records after approval. Rather than receiving instructions through chat, it operates autonomously in the background triggered by events — and since integration with SaaS is common in real work, this is a great reference.

In this hands-on session, AWS CDK is used to define infrastructure as code, deploying resources such as AgentCore Runtime, S3, DynamoDB, and Lambda all together. IaC is indispensable for production deployments, and being able to reuse infrastructure definitions created once in other projects is a major benefit. Since CDK fundamentals (such as the App / Stack / Construct hierarchy) are explained from scratch, even CDK beginners can approach it with confidence.

Chapter 16 is written from an organizational and business perspective on "how to introduce AI agents into business operations." Personally, this chapter hit home more than anything else.

In my own work supporting customers with AI agent adoption, it's not uncommon to provide support starting from a vague understanding of "what AI agents can actually do." This book introduces the idea of viewing AI agents not as automation tools like RPA, but as "a talented new hire." If you prepare business rules and procedures as prompts and documents, and set up access permissions to necessary tools, the agent will handle the work appropriately. This analogy is easy to understand and something I want to keep in mind at all times.

The book also repeatedly emphasizes the importance of "narrowing the scope and first showing something that works." Applications leveraging LLMs come with uncertainty, so trying to build a large-scale system all at once leads to frustration with gaps in output. Build something that works at 50% completeness first and show it to stakeholders, accumulate "small successes," get a feel for the key points of AI adoption, and then concretize the goals. This approach strongly resonates with me from my own experience.

Another point that struck me was that many cases of failed adoption are not due to technical issues, but to the problem of "nobody uses it." Integrating agents into users' daily workflows, making it possible to call agents from familiar UIs like Slack or email — it's truly sad when something you put effort into building goes unused...!

I really hope everyone reads this, engineers and non-engineers alike.

Who This Book Is Recommended For

• Those who want to build AI agents on AWS but don't know where to start
• Those who want to systematically learn Strands Agents and AgentCore
• Those who want to acquire knowledge that covers not just building AI agents but also operating them
• Those who want to learn by getting hands-on experience
• Those who are in a position to propose AI agent adoption to customers

The content spans a wide range from theoretical learning to actual code implementation, so while the main audience is engineers, the content is accessible enough that even non-engineers can dive in by leveraging generative AI. For those who have decided to build AI agents on AWS — definitely pick this up!

Closing Thoughts

There is no other Japanese-language book that covers the construction and operation of AI agents centered on feature-rich AgentCore in such a practical and systematic way. The structure — carefully explaining the basics of generative AI and each AgentCore feature in individual chapters, then combining them cross-sectionally in hands-on sessions — created a satisfying sense of understanding deepening step by step as I read. It served as a great review for me personally, and it's a book I want to be able to revisit anytime. Lately, I've been reading it every morning during my commute on the train.

2026 has the potential to be the inaugural year of AI agent development. For those thinking about introducing self-built AI agents into their work in this turbulent year, this is the book you should pick up first.
Let's use this book to get up to speed and dive headfirst into building AI agents with AgentCore!!!
I'll also keep sharing the latest information on AgentCore through my blog!

https://dev.classmethod.jp/author/yjinn/

Thank you so much for reading this article to the end! Nothing would make me happier than if it conveyed even a little of what makes this book so great!