How a website built with AI actually comes together in 2026 — from an idea to a finished, published site. What AI genuinely speeds up, and what still comes down to you.
The question I now hear more often than "what does it cost" goes roughly like this: "If AI writes the code, surely I'll just build the site myself over a weekend, right?". The short answer: AI really did speed up building a website — but it sped up the easy part, putting it together, not the decisions. Those aren't the same thing, and confusing the two is what costs the most. Why "putting it together" is the smaller part of the bill is something I covered separately in the cost breakdown of a website in 2026. Here I show what that faster way of working looks like in practice — from the first idea to a site you can open in a browser.
Most people picture it as a chat with a bot: you ask a question, you get a piece of an answer, you paste it in, something breaks, you ask again. That's the picture from a few years ago. Today sensible work looks different: a tool like Claude Code doesn't hand you scraps to glue together — it does the work itself. It looks over the whole project, proposes a plan, makes the changes, and immediately checks whether the site still works. And it does this on the real project, not in a window off to the side.
That difference is bigger than it sounds. In the "chat with a bot" version, you're the one gluing it all together, remembering what clashes with what, and catching that a fix in one place broke something three screens away. Here the tool takes that weight on itself and hands the result back in a form you simply look at: a ready preview of the site to click through. Your role shifts from doing to deciding and signing off — exactly the part you never wanted to hand off anyway.
The idea. There's no AI here — and that's the point. "I want a site that sells my course" isn't an idea yet, it's a wish. An idea is something concrete: what goes on the site and why exactly that. No tool will guess that for you — it doesn't know your market, your margins, or your customer.
The plan. Before anything gets built, you turn the idea into concrete decisions: what's on the site, what's deliberately left off, how it should work, and where it should "live". This is where the hard-to-undo decisions get made — a bit like the foundations of a house. You don't change these every month; you pick once and live with it for years, so you don't hand them to a tool in a single sentence. This plan is often short, but it steers everything else: the vaguer it is, the more rounds of fixes — and the more often you're fixing not the site, but your own unsaid assumptions.
Claude Code does its part. Only now does the tool come in: it takes the plan, proposes its own approach, and makes changes in small steps, checking each time that the whole still works.
You sign off. Every change is saved separately and waits for your "yes" before it goes further. This is where you take delivery of the work and catch what the machine lets through. The tool speeds up the putting-together; the judgement stays with a person.
Live preview. Every version gets its own link with a preview of the real site. You open it on your phone, send it to the client, you both look at the result before anything goes "for real". That's feedback measured in minutes, not in a back-and-forth of "please send a new version" emails. This part is handled by Vercel — the platform where the site ultimately lands, live.
Published. Going from a preview to a working site is one decision, not a new project. The whole route — idea → plan → AI builds → you sign off → preview → published — is one path, where the tool does the middle and the decisions stay on your side.
The most common misconception: since AI writes the code, the site is almost free. But putting the site together was never the bulk of the bill. If you break down the real time on a site built with AI, "building" is the only bar AI actually trims — and it sits next to several bigger ones it doesn't touch:
Decisions and checking together are well over half the effort — and both are, by definition, done by a person. AI lowers the "building" bar and helps a little with content, but checking what it did tends to grow rather than shrink: since it's produced faster, there's more to look over and consciously approve.
You can — really. Over a weekend you can go from nothing to a working, published site: the tool starts from the plan, follows it, shows preview after preview, and publishing is one decision. It's exactly the same path, just compressed into two days. With one caveat: fast doesn't mean hands-off — and that's where the part AI won't take off your shoulders begins.
“AI didn't build the site for you. It shortened the dullest part — the putting-together — and showed that the rest, the decisions and the sign-off, was always the real work.
”
