agents/prod
★
Resource · the reading path

Nine books, ordered — a path, not a pile

You want AI engineering, LLMs, and agents — with a focus on actually building things. This is that path: nine books, ordered. The point is the sequence, not the pile. Tick things off as you go, and use the filter to reorder it around how you like to learn.

One honest note up front: this field moves fast, and agent-specific content is still thin in book form — most of it lives in blog posts, framework documentation, and research papers. So treat stages 1–3 as the durable stuff and stage 4 as a scaffold you supplement with current writing (Anthropic's own posts on building effective agents and context engineering, plus framework docs, stay more current than any book can).

Choose your route

The order depends on whether you'd rather understand the internals first or start building immediately:

Interactive · pick your route

Build the projects alongside — not after

Reading nine books without shipping anything is the main failure mode here. One project per stage — tick them off (saved in your browser):

Interactive · project checklist
    Two rules that matter: build the eval harness before the agent, not after; and add logging/traces from day one — traces are what let you debug and improve agent behaviour.
    Tune this path to you

    Two questions sharpen it a lot. How much do you already code, and in what? Comfortable in Python → the from-scratch stage is genuinely doable, do it. Newer to programming → put building first and internals much later, or the from-scratch books will just stall you. And why are you learning? To get hired → weight evals, RAG and production reliability much heavier. To build a specific product → you can skip most of the from-scratch stage entirely.

    The one rule

    Ship something at every stage, and supplement the agent stage with current blog posts and framework docs — books lag that part by a year or more. The rest of this field guide is exactly that supplement.