Library · book

AI Engineering

Chip Huyen
2025·O'Reilly Media

Source: https://www.oreilly.com/library/view/ai-engineering/9781098166304/

The most product-facing book on this shelf.

Where the others explain how a model works, Huyen assumes you'll call one and asks the harder question for a team shipping software: what do you build around it? Evaluation, prompt engineering, retrieval, finetuning, inference cost — the disciplines that actually decide whether an AI feature is good.

For directing AI product, it maps almost exactly onto the decisions you own — build vs buy vs finetune, how to know it works, where latency and cost hide — from someone who has taught and shipped this at scale.

Central argument

A practitioner's map of the new AI stack: how engineering with foundation models differs from traditional ML, and then the disciplines that decide whether an AI product works — evaluation and benchmarking, prompt engineering, retrieval-augmented generation, finetuning, and inference optimization and cost. It treats the model as a component and concentrates on everything you build around it.

Critique

Because it starts from 'the model is a given', it deliberately says little about training or the internal mechanics — you won't learn how a transformer works here. It is also a fast-moving-field snapshot; the durable value is the framework (how to evaluate, how to choose, how to serve) rather than any specific tool.

Why it matters for product

For a CTO/CPO this is arguably the single most directly useful book on the shelf: it is organized around exactly the decisions you own — build vs buy vs finetune, how to evaluate reliably, where cost and latency come from, when RAG beats finetuning. It turns 'we should use AI' into a set of tractable engineering and product choices.