NEW 2026

Building AI Agents on AWS

by Stephen P. Thomas

A practical, hands-on guide to building production-ready AI agents using Amazon Bedrock, AgentCore, and multi-agent orchestration with AWS Step Functions. From your first Bedrock API call to deploying secure, cost-optimized multi-agent systems, this book covers the full AWS agentic AI stack — including RAG with Knowledge Bases, MCP integration, serverless AI pipelines, prompt engineering for AWS developers, and production observability. Every chapter builds toward real-world systems you can deploy with confidence.

What You'll Learn

• Understand why agentic AI changes everything and how the AWS agentic stack (Bedrock, AgentCore, Step Functions) fits together
• Make reliable Bedrock API calls with model selection, structured output, retry strategies, and cost planning
• Build autonomous agents with AgentCore including tool use, memory, and guardrails
• Orchestrate multi-agent systems with AWS Step Functions for complex workflows
• Implement RAG pipelines using Bedrock Knowledge Bases for grounded, accurate responses
• Secure AI agents with Cognito and IAM, and integrate MCP for standardized tool access
• Optimize costs with prompt caching, model routing, and deploy with full production observability

Full Table of Contents

1. 1 The Architecture of Intelligence: Why Agentic AI Changes Everything — Single API calls vs. agentic AI, the AWS agentic stack (Bedrock, AgentCore, Step Functions), Model Context Protocol, and environment setup
2. 2 Bedrock Fundamentals and Model Selection — How Bedrock requests flow, choosing models without guesswork, Converse vs. InvokeModel, reliability controls, security baselines, observability, cost planning, and the ClearPath Ticket Triage mini build
3. 3 Prompt Engineering for AWS Developers — Practical prompt engineering patterns tailored for AWS agentic workloads
4. 4 Building Agents with AgentCore — Hands-on agent construction using Amazon Bedrock AgentCore with tool use, memory, and guardrails
5. 5 Multi-Agent Orchestration with Step Functions — Designing and deploying multi-agent systems using AWS Step Functions for complex workflows
6. 6 RAG: Knowledge Bases on Bedrock — Building retrieval-augmented generation pipelines with Bedrock Knowledge Bases for grounded, accurate AI responses
7. 7 Securing AI Agents with Cognito and IAM — Authentication, authorization, and security patterns for production AI agent deployments
8. 8 Serverless AI Pipelines (Lambda + Bedrock) — Building event-driven, serverless AI processing pipelines with Lambda and Bedrock
9. 9 MCP Integration for AWS Workflows — Implementing Model Context Protocol for standardized tool integration across AWS services
10. 10 Cost Optimization and Prompt Caching — Reducing AI costs with prompt caching, model routing, token optimization, and usage monitoring
11. 11 Production Deployment and Observability — Deploying AI agents to production with monitoring, tracing, alerting, and operational excellence
12. 12 About the Author
13. A Review Question Answer Key — Answers for all chapter review questions covering Chapters 1–11