Acknowledgments

This book wouldn’t exist without the engineers who shared their experiences - the successes, the failures, and the hard-won lessons that no documentation captures. The patterns in this book emerged from hundreds of conversations, code reviews, and post-mortems with teams building agent infrastructure in production.

Thanks to the open-source community that built the tools this book describes. The teams behind Ona, OpenFGA, LangChain, CrewAI, AutoGen, and the Model Context Protocol created the infrastructure that made agent engineering a discipline rather than an experiment. Their willingness to build in the open, document their decisions, and accept feedback from the community accelerated the entire field.

Thanks to the researchers whose work underpins every chapter - the teams at Stanford, UC Berkeley, Google, Anthropic, OpenAI, and dozens of universities who published the papers that turned intuitions into evidence. Special thanks to the authors of “Lost in the Middle,” “Zanzibar,” and “MCPMark” - papers that changed how I think about context, authorization, and evaluation respectively.

Thanks to the early readers who caught errors, challenged assumptions, and pushed for clarity. Technical books are only as good as their reviewers, and this one benefited enormously from engineers who read drafts and said “that’s not how it works in practice.”

Finally, thanks to every engineer who has ever stared at an agent trace wondering why it made that decision. Your frustration is the reason this book exists. I hope it helps.