Books and resources
Eight things worth reading, each with the chapter where you should stop. Nothing here is sponsored, and there are no affiliate links on this page.
Long form
Books
- AI Engineering: Building Applications with Foundation Models ↗Chip Huyen · O'ReillyFor: Anyone shipping a product on top of a model rather than training one.Why: The closest thing to a field manual for this work: evaluation, retrieval, prompting, and the operations around them, written for engineers who will be on call. It treats evals as the center of the discipline, which is the correct emphasis and the one most tutorials skip.Where to stop: If you only read two chapters, read evaluation and RAG. The rest is context you can pick up from docs.
- Prompt Engineering for LLMs ↗John Berryman and Albert Ziegler · O'ReillyFor: People who write prompts for a living and are tired of tips.Why: Written by people who build the tooling, so it explains why a prompt behaves the way it does instead of offering incantations. The sections on structured output and evaluation of prompts are the parts you will return to.Where to stop: Skip the introductory model theory if you have used an API before. Start at structured output.
- Hands-On Large Language Models ↗Jay Alammar and Maarten Grootendorst · O'ReillyFor: Anyone who wants the mental model of how these systems work, visually.Why: Alammar's diagrams are the reason this book exists. Attention, embeddings, and fine-tuning are explained with pictures before equations, which makes it the fastest honest route from user to informed builder.Where to stop: Read the first half. Fine-tuning chapters matter only if you are actually fine-tuning.
- Build a Large Language Model (From Scratch) ↗Sebastian Raschka · ManningFor: Engineers who want to build a toy model end to end to understand the real ones.Why: You implement attention, tokenisation, training, and sampling in code. It is the difference between knowing that models have a context window and knowing why truncation behaves the way it does.Where to stop: This is optional. If your job is building on models rather than training them, read AI Engineering first and come back only if curious.
Reference
Ongoing reading
- Model Context Protocol specification ↗AnthropicFor: Anyone writing or debugging an MCP server.Why: The spec is short and readable, and it settles most arguments. When a client behaves oddly, the answer is usually in the transport or lifecycle section rather than in a blog post.Where to stop: Read the lifecycle and transport pages first. The rest is reference.
- Building effective agents ↗Anthropic engineeringFor: Anyone about to over-engineer an agent.Why: The clear statement of the boring truth: most tasks need a workflow with one model call, not an autonomous loop. Reading it before building saves a month.Where to stop: It is one sitting. Read all of it.
- OpenAI Cookbook ↗OpenAIFor: People who learn by copying working code.Why: Runnable notebooks for the API surface, kept current. Useful when the documentation tells you what a parameter does and you want to see it used.Where to stop: Search it rather than reading it. It is a reference, not a course.
- Simon Willison's weblog ↗Simon WillisonFor: Anyone who wants to know what changed this week and whether it works.Why: The most consistently useful running commentary on LLM tooling, written by someone who actually tries things. The MCP coverage in particular is more practical than most vendor documentation.Where to stop: Follow the LLM tag. Skip everything else unless it interests you.
How these were chosen
Only things someone on this desk has actually read or used. The "where to stop" line is the point: most technical books are 40% longer than the value they hold for a working developer, and saying where to stop is more useful than a rating.
No affiliate links, no publisher relationships. If a link rots, tell us and it gets replaced or removed.