Use Claude like you mean it.

One place to chat, create, and open your work — claude.ai on the web, the desktop app, or the phone. Same account, same history, everywhere. Learn the home base once and everything else clicks into place.

Learn how the engine works, and everything it does stops feeling like magic — including the mistakes. This is the chapter that keeps you out of trouble.

Where your data lives — the three levels

Claude only knows what it's given. Knowing which level you're in tells you how reliable and how private an answer is. And the rule that never changes: permissions inherit — Claude can only ever see what you can see.

The work order: G-C-E-S

Think of a prompt as a work order. A strong one has four parts — miss one and you get a weaker answer. This framework is ours, and it works with any AI tool.

Stop regenerating answers. Start building things you keep. An artifact is a real thing Claude builds — a dashboard, a calculator, a document — that lives beside the chat and edits in place. You don't regenerate; you refine: "make negative variance red," "add a total row."

Projects — give Claude your world once

A Project is a container: your documents, your standing instructions, and its own chat history in one scoped space. Open it and Claude already knows your world — no re-explaining. One project per workflow — close, AP, reporting — not one giant catch-all.

Where an instruction belongs — the four levels

Everything you just learned stacks into four levels, and one rule runs all of them: pick the lowest level that still covers it — and the moment you're typing the same thing again, move it up. Each level sticks harder than the one below it, so the higher you go, the less you repeat yourself.

A skill captures how a job should be done — the steps, the format, the rules — so anyone can trigger it in plain language and get a consistent, professional result. You don't code a skill. You describe it.

So far Claude only sees what you paste. Connectors change that — a secure door into the tools where your work already lives. This is the leap from "AI that reads what I paste" to "AI that works in my actual filing cabinet."

What "agent" actually means — the honest scale

"Agent" gets thrown around loosely. Here's the real spectrum — and the one golden rule that holds at every level: Claude gathers and drafts. A human approves. Never auto-post anything irreversible.

The built agent you already have: Cowork

Chat answers you; Cowork does the work. It's the level-3 agent included with every paid plan — Pro, Max, Team, and Enterprise — and you find it right next to Chat. Give it a goal and it plans the steps, runs them end to end, and hands you finished work to review. It runs on web, desktop, and mobile (mobile in beta); the desktop app is what reaches the folders and applications on your own computer.

Three real workflows — not toy examples

Actual jobs, start to finish. The pattern in all three: Claude does the gather-and-first-pass; you keep the judgment on every exception.

Chat answers you. Cowork does the work. Past those sit three more surfaces included in the same paid plan — and most people never open a single one. They aren't add-ons and they aren't extra money: Code works on the files on your machine, Chrome works inside your browser, and Design turns a rough idea into a finished deck. One of the three isn't for developers at all.

Claude Code — the one that edits real files

Anthropic's own description: an agentic coding tool that reads your codebase, edits files, and runs commands. The honest test of whether it's yours: does your work live in files on a computer, or in a browser and a mailbox? Files — keep reading. Otherwise, skip ahead to Chrome; you're not missing anything.

Claude in Chrome — the one that clicks for you

A browser extension that lets Claude read, click, and navigate websites alongside you. Included with every paid plan. It's the most immediately useful of the three for non-technical work — and the one that demands the most judgment, because Anthropic's own documentation says it plainly: it is still risky. Teach it with that sentence attached.

Claude Design — the one nobody expects

The surprise of the three. It isn't a coding tool and it isn't a chat window: you describe what you want and Claude builds the actual visual — a deck, a prototype, a landing page, a one-pager. It's in beta on Pro, Max, Team, and Enterprise. One catch worth knowing before you go looking for it: Enterprise has it switched off by default.

So which one is actually yours?

Most people need one of the three, not all three. Match it to where your work already happens and ignore the rest until that changes.

Chapter 4 taught you skills — one procedure, packaged once. A plugin goes further: it's a box that can hold many skills, custom agents, event hooks, and live connections, installed together with one command. A skill is a procedure. A plugin is a department.

The part nobody teaches honestly. Claude comes in a small family named after forms of writing — the bigger the form, the bigger the model. Pick by the task, not by habit: you don't need the biggest one for most work.

Where the money actually goes

A person chatting on a paid plan is a flat monthly fee. The cost surprises come from machines acting on their own. Know the difference and you'll never get a shock bill. (Plans run roughly free → $20 → $100+ per month, per-seat for teams — numbers move, so verify at the source before quoting.)

Running it for a team — the owner's corner

When it's the whole office and not just you, four things keep the bill boring and the setup clean. This is the governance your leadership will ask you about.