You've used ChatGPT. Now learn to run it.
You don't need a bootcamp. You need an afternoon, the official free courses, and an order to take them in. That's this page. Then the book.
Stop using AI like a chatbot. Start using it like an OS.
The full reframe is the prologue. Read it →
Jump to section tap to open
Is this you?
You opened ChatGPT, asked it a few things, and closed the tab. It was fine. It didn't change how you work. That's because you used it like a chatbot — a thing you visit. This page is the other door.
Two ways in. Pick yours. The page serves both — it splits at the right moment and meets back up at the end.
You run a business, a team, or your own calendar. You don't want a terminal. You want Claude doing real work — drafting, sorting, summarizing, running a job every morning so it's done before you sit down. You'll never write a line of code. You'll still ship things that run without you.
You're comfortable in an editor and a command line. You want Claude inside your repo — reading the codebase, planning a change, writing it, committing it. You want subagents, connectors, the API when it's time to ship software. This track gets you there.
The full who-and-why — who this book is for and who it isn't — lives in the prologue. Read it → Then come back and start the ladder.
Five words you need first
Five words show up everywhere from here on. Learn them now, in one breath each. These are the beginner-grade lines — the full glossary has the depth.
The brain. Claude is a model. So is the thing under ChatGPT. It reads text and predicts what comes next, trained on a huge amount of writing. You don't manage the model — you direct it. glossary →
A window vs a worker. Chat answers and waits. An agent takes actions — runs tools, reads and writes your files, loops until the job is done, then comes back. The whole jump in this book is from chat to agent. glossary →
The agent in your terminal. It works inside your files and your code: explore, plan, write, commit. This is the builder's surface. glossary →
The agent in your browser. Same worker, no terminal. It runs a task, writes the file, comes back. This is the operator's surface — Claude with a desk and a filing cabinet. glossary →
The standard plug. It's how an agent reaches your tools and data — your calendar, your files, a database. Plug it in once; the agent can use it from then on. glossary →
Want the deeper version of any of these? The vault and connectors get a full chapter → Ch 4 — the vault. Otherwise, keep going — next is the order to take it all in.
The learning ladder
Four rungs. Every rung tells you the level, the why, the time, and whether it's mandatory or optional. Nothing here is a wall of forty links. Take them in order.
- 1Get the conceptsEveryone · Mandatory · You just did it
The five words above and the reframe. That's the floor. You're standing on it now.
- 2Take the official coursesEveryone · Mandatory · ~1 hour to start
The free, official Anthropic courses, in order. One course everyone takes; then one track-specific course; then your first reusable Skill. This is the rung that does the real teaching. Jump to the courses →
- 3Orient, then set upEveryone · Mandatory · self-paced
Read why this matters → how to read this book. Then set up your machine — twelve concrete steps, install to first scheduled task → day zero. Builders do the hands-on here. The ladder routes you to the setup; it doesn't repeat it.
- 4Read and buildEveryone · The point of all of it
Open chapter 1 and build alongside it → Ch 1 — the day I killed my tabs. Courses end. Building starts. Ship one thing that runs without you.
That's the climb. Concepts, courses, setup, build. Start the courses now.
Official courses — Anthropic first
These are free, on Anthropic's own domain, and ordered. No third-party rehashes — go straight to the people who built the tools. Each course earns its slot: a reason, a level, a track.
Do these first — the foundation. Everyone starts with Claude 101. Free, email signup, no card. By the end you've used Claude to do one piece of real work — not asked it a trivia question. That's the line. Cross it here. Then take the course for your track: Introduction to Claude Cowork if you're operating no-code, Claude Code 101 if you're building. One catch on Claude Code 101 — the course is free, but the hands-on exercises want a Pro/Max plan or an API key. Have one ready before you start.
Use Claude for real work — writing, data, documents, projects.
Install to daily workflow — the Explore, Plan, Code, Commit loop.
↳ Hands-on exercises need a Pro/Max plan or API key.
First launch to daily use — task loops, skills, real file output.
↳ Cowork runs on a paid Claude plan.
Go deeper with Claude. Both tracks take Introduction to Agent Skills next — Skills are markdown, no code, the first reusable thing you'll own. Builders keep going: Claude Code in Action, Introduction to Subagents, then Building with the Claude API and Model Context Protocol when you need them. The API and MCP courses need Python — take them when the work calls for it, not all at once. The docs and cookbooks sit alongside as the manuals: reach for them, don't read them cover to cover.
Go deeper with Claude — 9 more (courses, docs, cookbooks) tap to open
Write reusable markdown Skills Claude auto-applies. No coding.
Drive Claude Code inside a real development process.
Delegate work and manage context by splitting tasks across subagents.
API auth, prompting, structured output, tool use, RAG, caching.
↳ Requires Python proficiency.
Build MCP servers and clients in Python — tools, resources, prompts.
↳ Requires Python.
Official Claude Code docs — install, CLAUDE.md, hooks, MCP, subagents.
Official platform docs — API, models, prompting, agents.
Official code recipes — runnable notebooks for API, tools, RAG.
The open MCP spec — the standard behind connectors.
↳ Open standard, not on anthropic.com.
Take Claude 101, then the one for your track. Then write your first Skill. That's operator-grade.
Going further
Most people arrive knowing one tool — usually ChatGPT. If that's you, these are the concept on-ramps. They teach you what a model and a prompt are; then you come back to Claude for the rest of the ladder. Take one path to the reframe, not all of them.
If you live in OpenAI's world: start with Getting started with ChatGPT and Prompting fundamentals on OpenAI Academy. Concept-equivalent to Claude 101 — then converge back here.
If you live in Google's world: open Google AI Studio, type a prompt, set a system instruction, hit Run. Then the 45-minute Introduction to Generative AI badge on Google Skills. Same idea, different stack.
Want the under-the-hood version: Andrew Ng's Generative AI for Everyone and AI Prompting for Everyone on DeepLearning.AI explain how this works without code — the courses are watchable free; the graded certificate is paid. Hugging Face's AI Agents Course goes hands-on if you've got basic Python. These are independent of the labs — legit, but not on a toolmaker's own domain. Flagged as such on the cards.
Other ecosystems — OpenAI & Google (9) tap to open
Your first conversation — write, brainstorm, solve problems.
Write clear, effective prompts — the core everyday skill.
OpenAI's free learning hub — videos, guides, live events.
Your first API call — keys, SDK, text, tools, streaming.
↳ API usage needs paid credits.
Runnable recipes — agents, evals, RAG, function calling.
Browser playground — type a prompt, set a system instruction, Run.
45-minute primer on what generative AI actually is. Free badge.
Free API key, install the SDK, first call in under five minutes.
Full developer reference — text, tools, function calling.
Hand-picked extras — DeepLearning.AI, Hugging Face, Microsoft (7) tap to open
Andrew Ng teaches non-technical people to prompt ChatGPT and Claude.
↳ Watch free; graded certificate is paid.
Andrew Ng: how generative AI works and where to use it. No coding.
↳ Watch free; certificate is paid.
90-minute look under the hood: tokens, attention, prediction.
↳ Watch free; certificate is paid.
Anthropic's own course on driving Claude Code across a real codebase.
↳ Free during beta; certificate is paid.
Anthropic's Elie Schoppik shows you how to build and deploy an MCP server.
↳ Free during beta; certificate is paid.
Free course, free certificate — what agents are and how to build them.
↳ Hands-on units need basic Python.
21 free lessons covering prompts, RAG, agents, and more.
↳ Assumes basic Python or TypeScript.
One thing this section is not: a template library. If you came looking for CLAUDE.md skeletons, hooks, or Skill files to copy-paste, those aren't courses — templates to copy, not courses →
What you'll be able to do
This is the payoff. Not theory — things people shipped with this exact stack. Each one is a rung on the same ladder you're climbing.
An idea becomes a company artifact — in a weekend
A wild pitch sat in someone's head for weeks. One Saturday, two linked interactive HTML documents shipped — a deck and a profit calculator — and the artifact existed before the next meeting did. That's "I have an idea and a real thing now." →
A real product, shipped in a Saturday
An idea over coffee Saturday at 9 AM; four hours of actual hands-on work spread across a weekend of brunch and walks; by Sunday night a real URL, a real cron, and a 90-second audio brief that plays on the walk. $80 in tokens, zero salary — a senior engineer would've billed $2,000 and shipped in a week. That's "I shipped." →
A refactor across twelve files — the swarm holds it in its head
Work that used to eat a sprint: a dynamic workflow planned the decomposition, fanned out across the files in parallel, ran verifier passes to catch its own mistakes, and came back with a diff you could read top to bottom. You still read every line. That's "I scaled it." →
A whole company runs on it
Satispay — Italy's leading fintech — now writes 75%+ of its code with Claude. And Lovable, built on Claude, lets anyone create software 20x faster. That's "my operation runs on this." →
That's the progression — idea, ship, scale, operation. You start at rung one. These are where it goes.
Your next 3 clicks
Courses gave you the tools. The book is the keys. Three clicks, in order.
Read the prologue. Who this is for, the reframe in full, the six-part map. Fifteen minutes.
2Twelve steps, clean machine to first scheduled task. Tick the boxes. You'll have something running in your name before the coffee's cold.
3The day I killed my tabs. Read it with your terminal open and do the thing it describes.
Stuck? Install won't run, "command not found," no API key — it happens to everyone on day one. Paste the exact error back into Claude and ask it to walk you through the fix, one step at a time. The tool that's confusing you is also the best support agent you have.
You've crossed the line — AI as an OS, not a chatbot. Now pick one job you do every week and hand it to Claude. Ship it. Watch it run once without you. That's the whole move. Turn the page.
The whole arc is 45 chapters. Start with one. → the journey