It’s a magic power. You just have to learn to use it.
You’ve already talked to AI. Now watch it build you something real — try it right here, no sign-up.
no card tricks. real software, from one sentence.You’ve been using AI like a smarter Google. That’s not the magic.
Chatting gets you answers. But the moment it stops talking and starts making — a real tool, a working page, a thing you can actually use — something clicks. You stop asking “what should I do?” and start saying “build me this.”
“Can you explain how to make a budget?”
→ three paragraphs to read“Build me a budget for my month.”
→ a working tool, made for youThat’s the whole shift. Everything below is just getting you there — for real, on your own machine.
Now let’s get you set up to do this yourself.
The thing you just tried up top? You’re about to do that on your own machine, where what you build is really yours.
The building part costs money: Claude Pro, about $20 a month. You will hit that as soon as you open the Code tab, so we would rather you heard it here than met it as a surprise halfway through.
The free plan can chat, but not build. That is exactly why the demo at the top of this page is free: so you could see what this is worth before spending anything. If you are not ready to pay, stop after step 1 and read on. There are free routes further down the page and nothing below this point breaks without a subscription.
Download the Claude desktop app
Clicking below opens Anthropic’s download page in a new tab. Pick Mac or Windows, and it installs like any normal app. No terminal, no setup wizard from hell.
Open the download page ↗Sign in, and open the Code tab
This is where the building happens. One honest heads-up below before you go further.
Make an empty folder, call it my-first-project
Put it anywhere — your Desktop is fine. When Claude asks which folder to work in, pick that one. (This trips everyone up: Claude just needs a home for what it builds. An empty folder you made on purpose = nothing to worry about.)
If any of that felt confusing, that is on the instructions, not on you. This is the step people get stuck on most, every time. Read it again slowly, or skip ahead and come back to it. Nothing further down depends on you having finished this first.
The terminal
Same Claude. Same subscription — nothing more to buy. It just stops having a limit. You never have to come through this door: every step in this guide works in the desktop app, exactly as described above.
- ◆It remembers how you work
You write one plain-English file —
CLAUDE.md— saying how you like to be worked with. Every project, every time, forever. No re-explaining yourself. - ◆It works on your whole machine
Not one folder you picked in advance. Anything you point it at.
- ◆Many things at once
Several sessions running side by side, each on its own piece of work.
- ◆No ceiling
Everything above is where the terminal starts, not where it stops.
Nothing here can break your computer. Once Claude is running, it asks before it changes anything — you read it, you say yes. Nothing happens that you didn’t approve. (Installing the terminal version is the one exception: that step just runs, like any app installer.)
This time, you make the thing.
Four small moves. No guessing what to type — we hand you the exact words. Click through them here first, then do it for real in the app.
Point Claude at your folder
In the Claude app, open the Code tab and click “select folder.” Choose the empty my-first-project folder you just made. That’s Claude’s workspace now — everything it builds lands there.
There’s really only one skill — and you just used it.
The cleverness is already in the machine. The only thing that decides whether you get magic or mush is how you talk to it — and that’s learnable. Everything else is just this loop, over and over:
…then around again. Two or three laps and the thing is exactly what you pictured.
Three moves that make the loop sing
Use plain, messy words
There’s no secret syntax and no “one perfect prompt.” Talk to it like a clever friend who can’t see your screen. “Make the text bigger and the blue softer” works perfectly.
When you’re stuck, ask it what to say
Blank on what to type next? Hand the problem back: “what else should I tell you to make this better?” It’ll ask you the right questions.
Ask it to critique its own work
Before you settle, try: “what could be better here?” It’ll spot things you wouldn’t and offer to fix them. The first answer is never the best one.
A tiny recipe for a prompt that lands
Most weak prompts are missing one of three things. Add all three and the answer gets useful fast.
Like this: “I run a tiny bakery [situation] — make me a one-page menu I can print [goal] — keep it to six items and warm colors [constraint].”
You can approach almost anything now.
The loop you just learned isn’t only for web pages. It’s for almost any small thing you’ve been meaning to make. Pick a corner of your life — here are real first prompts to steal.
Build me a simple one-page website about what I do — a short intro, three things I offer, and a way to get in touch. Keep it clean and calm.Make me a checklist tool for [the task I repeat every week] — let me tick things off, and start fresh each week.Here are my messy notes: [paste them]. Turn them into a short, clear summary I could send my team — bullet points, no fluff.Copy one, open the Code tab, paste, and run the loop. Then — the real graduation — write your own. You already know how: plain words, look, say what’s off, repeat.
It really is a magic power. You just had to learn to use it — and now you have.