Part I: One Agent

Every system in this book, is built out of one repeating unit: a single agent doing a single task well. Teams that skip this level and jump straight to orchestration inherit every weakness of their weakest agent, multiplied by the number of agents. So the walkthrough starts here.

This part gives you the working vocabulary and the craft for that unit:

• What an Agent Is defines the thing precisely: a model plus the machinery around it, and the five levers that determine whether it works.
• Instructing an Agent covers prompting as an engineering discipline rather than a knack: structure, language, and the research on what measurably matters.
• Working With Models explains the substrate: why models are probabilistic, how errors compound across steps, and how to choose one.
• Giving an Agent Tools covers the action surface: designing, describing, and restricting what an agent can do.
• The Agent Loop puts it together: the think-act-observe cycle inside every agent, and the autonomous outer loop that turns one agent into a worker you can leave alone.

If you already run agents daily, skim this part for the quantified findings (they are cited, and several are counter-intuitive) and rejoin the walkthrough at Part II, where reliability engineering starts.