Agent Engineering using Claude

Agent Engineering is a practical guide to designing, building, evaluating, and operating reliable AI agents with Claude.

Most material on agents focuses on prompts, frameworks, or impressive demonstrations. This book focuses on the harder engineering questions: When should you use an agent? When is deterministic software better? How do you control tool use, manage context, evaluate behavior, handle failures, introduce human approval, and move from a prototype to a production-grade system?

The book is organized as a progressive journey from first principles to production operation. Readers build a working reference implementation—the Claude SDR Lab—while learning the engineering decisions behind it.

The first part, Foundations, covers:

• What makes a system an agent
• When agents are—and are not—the right architectural choice
• Designing agent workflows before writing code
• Giving Claude business context and operating instructions
• Structuring tools, prompts, state, and control flow
• Building a minimum viable agent
• Testing and evaluating agent behavior
• Adding human review and approval
• Improving reliability, cost, and predictability
• Preparing an agent for real-world deployment.

Later parts expand into memory, retrieval-augmented generation, planning, multi-agent collaboration, resilience, security, guardrails, AgentOps, observability, scaling, multimodal agents, knowledge graphs, and model adaptation.

This book is written for software engineers, architects, technical leaders, consultants, and product builders who want to move beyond agent demos and learn how to engineer systems that can be trusted.

This is an actively developed early-access edition. The initial release contains the complete Foundations section. New chapters, examples, lab exercises, corrections, and implementation updates will be added periodically. Your purchase includes access to future versions of this edition.