Building, Evaluating, and Operating Reliable AI Agents
A practical guide to the systems around an AI agent: execution loops, tool interfaces, permissions, context, evaluation, and operations. Follow the ideas through runnable Python examples in the companion code.
Available from Amazon , AbeBooks (hardcover) , Orell Füssli (hardcover, Switzerland) , ThriftBooks (hardcover) , and YES24 (hardcover, South Korea)
The model is one part of an agent. The harness determines what it can do, what evidence it sees, when it stops, and how its work is checked.
Starting with a small execution loop, the book develops the engineering decisions behind reliable agent systems. Examples cover tool contracts, permissions, controlled edits, context, checkpoints, retries, security boundaries, tracing, evaluation, resource budgets, coordination, and release changes.
The companion contains runnable Python checkpoints, tests, and saved evidence. Start with the offline Chapter 2 demo, then follow the chapter instructions as each mechanism is introduced.
Browse the companion on GitHub ↗