Introduction
This is a book for clinicians who want to learn to build software, written by a clinician who learned to build software. It is not a computer science textbook. It is not aimed at engineers. It is aimed at the pediatric oncologist who wonders if she could automate the pathology summary she writes every Tuesday. It is aimed at the clinical pharmacist who is tired of explaining the vancomycin dosing rule to every new resident and wants the rule to live somewhere other than his head. It is aimed at the nurse manager who has three Excel spreadsheets that should really be one dashboard.
If those people sound like you, you are the audience.
What this book does and does not assume
It assumes you can read English at a professional level. It assumes you can install software on your laptop with administrator privileges. It assumes you have access to an Anthropic account (or are willing to make one — pricing is covered in Chapter 7) and roughly four to six hours of patience.
It does not assume you know what a terminal is. Chapter 0 explains that.
It does not assume you know what KHCC’s AI Office builds. Chapter 0.5 explains that.
It does not assume you have ever written code in your life, although if you have, you will move faster.
How the book is organized
There are six parts, twenty-four chapters, and two appendices.
- Part 0 — Foundations. What a terminal is (Chapter 0) and what the AI Office builds (Chapter 0.5). Two short chapters, written so a clinician with zero prior exposure can get unstuck.
- Part 1 — Getting Started with Claude Code. What Claude Code is, how to install it, how to talk to it, and how to keep it from doing anything you’ll regret.
- Part 2 — Daily Workflows. Your daily flow: files, context, money, git, and how to make Claude think before it codes.
- Part 3 — Making Claude Code Yours. The customization layer —
CLAUDE.md, slash commands, skills, subagents, and hooks. This is where Claude Code goes from useful to indispensable. - Part 4 — Pushing the Limits. Long-running and autonomous patterns: MCP servers, parallel worktrees, scripted runs, and overnight Ralph loops.
- Part 5 — The Part of Tens. Four short chapters of practical lists: ten clinical workflows, ten rookie mistakes, ten slash commands worth memorizing, ten lines for your
CLAUDE.md.
Two appendices close the book: a one-page cheat sheet, and a short tour of the Claude Agent SDK for readers who want to embed Claude in a Python or Node service later.
The four margin icons
Throughout the book, four kinds of paragraph are flagged with an icon in the margin.
💡 Tip. A shortcut, a keystroke, or a settings tweak that will save you time. Read these even when you’re skimming.
🧠 Remember. A concept you’ll use repeatedly. If you forget the rest of the chapter, hold onto these.
⚠️ Warning. Things that have cost real clinicians real money, real data, or real sleep. Read every single one.
🔧 Technical Stuff. Optional plumbing. Safe to skip on first read; useful when you come back to debug.
A word about the examples
Every concrete example in this book comes from a real AI Office pipeline at King Hussein Cancer Center. Pathology extraction is real. The AKI alerter is real. The Cockcroft-Gault calculator is real — Chapter 0.5 introduces it, and it is small enough that once you have finished Part 1 you can build a working copy yourself, from a blank folder to a running web app, in a single Claude Code session.
We use these examples instead of generic “build me a todo app” demos for two reasons. First, they are the actual work this book exists to support. Second, they are far better at teaching the judgment part of clinical AI than abstract examples can be — which fields to extract, which to skip, what counts as a regression, how to handle ambiguity. That judgment is most of the value a clinician brings to the work.
How to read this book
If you have never opened a terminal in your life, read Part 0 carefully, then read Chapters 1 and 2 carefully, then skim the rest until you have a project of your own to apply it to.
If you have used the terminal and dabbled in Python, skim Part 0 and start at Chapter 1. Come back to Part 0 when you need to explain it to a colleague.
If you came here only to learn how to build long-running autonomous agents (overnight refactors, eval-gated nightly runs), read Chapters 1, 4, 10, 14, and 18, then Appendix B.
If you just want the shortcuts, flip to Appendix A, then come back when something stops making sense.
A word from the editor
The first version of this book was written in a wry, Silicon-Valley-coder voice. It was readable, it was funny, and it assumed too much. A clinician reviewing the draft pointed out, fairly, that you cannot tell someone what a worktree is when you have not yet told them what a directory is.
This edition fixes that. The voice is calmer. The vocabulary is defined as it appears. Where the original waved at concepts and moved on, this edition stops and explains. Where the original assumed you already knew what aidi_catalog was, this edition has a whole chapter (0.5) explaining what the AI Office builds and why.
The earlier version still exists for readers who came in with technical fluency. This one is for the doctors, nurses, and pharmacists who are at the front of the queue.
Welcome to it.