Preface

Why this book

AI has moved from research labs into everyday products faster than most people’s understanding of it has kept up. That leaves a gap: you can use these systems fluently — prompt them, integrate them, make decisions about them — while still not having a working model of what’s actually happening underneath. That gap is uncomfortable for anyone whose job depends on judging what these systems can and can’t do.

The usual explanations don’t close it. One kind treats the model as magic — a black box that “just knows things” — which leaves you with no way to predict when it will fail. The other kind starts from equations and implementation details, which is precise but assumes the exact background most readers don’t have and don’t need. This book tries for a third path: build a durable mental model — tokens, embeddings, attention, training, retrieval, tool use, reasoning — using nothing but prose, worked examples, and precise language, so that the understanding still holds up after the specific models and products in use today are replaced by whatever comes next.

How this book came to be

This book started as personal notes — explanations, analogies, and worked examples accumulated over years working professionally in AI, originally written to clarify my own thinking and to help colleagues and non-technical stakeholders actually understand the systems we were building. Organizing those notes into a coherent, chapter-by-chapter manuscript was done in collaboration with Claude (Anthropic), which did the writing itself, and ChatGPT (OpenAI) used for editorial review along the way.

Who this is for

Managers, consultants, product leaders, founders, sales professionals, executives, students, and lifelong learners — anyone who wants to read an AI engineering blog or product announcement and actually evaluate it, rather than just trust or dismiss it.

What you’ll come away with

By the end, you should be able to explain how large language models generate text; why they hallucinate; what tokens, embeddings, attention, retrieval-augmented generation, vector databases, tool calling, the Model Context Protocol, and AI agents actually are; and — most importantly — tell a genuine advance apart from a marketing headline.

How the book is built

Every chapter builds understanding through a real-world story, a second worked example from a different angle, and a precise technical explanation, addresses the misconceptions that trip readers up, and ends with one bolded, memorable sentence that distills its core idea. Each chapter approaches its central idea from more than one direction before moving on, entirely through prose and carefully reused analogies — no diagrams to lean on instead of understanding the mechanism itself.

How to read it

Chapters are meant to be read in order, not sampled topically. Each chapter’s vocabulary is a prerequisite for the next, following an explicit concept dependency graph rather than a loose topic list — by the time a term like “attention” or “fine-tuning” is used without explanation, it’s because an earlier chapter already built it from scratch.