How to build a website with AI, end to end

The honest version first

AI can build a website fast, but it cannot decide what the website is for. The parts it does brilliantly — layout, components, first drafts of copy, wiring pages together — are the parts that used to be slow. The parts that still need you — what is true about you, what a visitor should do, whether it feels right — are the parts that were always the real work. This walkthrough keeps those two things separate, so you know where to lean back and where to lean in.

Step 1 — Write the brief (you decide)

Before any building, answer four questions in plain sentences: who is this for, what should they do, what is true about you, and what feeling should the page carry. Three or four lines each is enough. This is the one step you cannot delegate — everything the AI makes downstream is only as honest as this brief. Vague brief, generic site.

Step 2 — Agree the structure (AI proposes, you approve)

Ask the AI to turn the brief into a page outline: what pages exist, what each one says, and in what order. Read it as a sitemap, not as code. This is a decision point — cutting a page here costs a sentence; cutting it after it is built costs an argument with a generated layout. Approve the shape before a single section is designed.

Step 3 — Let it build (AI works)

Now hand it off. The AI builds the actual pages — structure, styling, components, placeholder-free content drawn from your brief. On The Cloud, this is a Studio app: a real project with real files that renders at its own URL the moment it exists, with nothing to install on your machine. You describe, it writes the files, and you have something to look at instead of imagine.

Step 4 — Review against the truth (you decide)

Open the live URL and read it as a stranger would. Two passes: first for honesty — is every claim true, is nothing invented, does it say what you actually do — then for feel. Mark what is wrong in plain language. “This headline overpromises,” “this section is filler,” “the tone is too stiff.” You are not rewriting; you are directing the next pass.

Step 5 — Iterate in tight loops (AI works)

Feed your notes back and let the AI revise. Good tools close this loop without you: The Cloud renders an app headlessly — desktop and mobile screenshots, plus console errors — so the AI can see what it shipped, fix it, and rebuild. Keep the loops small. One clear instruction, one look, one correction beats a long wishlist the AI half-satisfies.

Step 6 — Publish (you decide, deliberately)

Going live is a decision, not an accident. On The Cloud, making an app public waits until you confirm it in the app, so nothing you are drafting ships on its own. Public sites can carry their own link-preview cards and serve on a custom domain. Press publish when the site is true and reads right — not before.

Keep it current after launch

A website is never finished. Recurring updates — a new testimonial, a changed price, a fresh section — can run as a scheduled agent against the site, with its own write policy and budget, so small edits happen without you opening an editor. You still approve what goes live; you just stop being the one who remembers to do it.

Start here

Write your four-line brief, open The Cloud, and describe the site to Ora — the workspace opens for everyone, immediately, and the first draft comes back at a live URL. Or connect the AI you already use over MCP and ask it to build the site there. Either way, you spend your time deciding, and the machine spends its time building.

Frequently asked questions