Not prompts, but loops: how your website actually gets built

In our last piece on the builder we talked about what goes into the AI — the brand, the audience, the commercial goal, the standard of “good” — and called that craft context engineering. This piece is about the other half: how it runs once that context is in. Because the honest answer is that we don’t sit and prompt the AI a site into existence one instruction at a time. We build loops, and the loops do the building.

A prompt asks once. A loop keeps going.

Most people picture working with AI as a conversation: you type a request, you read what comes back, you decide whether it’s any good, you ask again. You are the judge of every single turn, by hand. That’s a prompt — a single ask, done manually.

A loop is a different thing entirely. Instead of judging each result yourself, you build a small system that does the asking and the judging on its own: it produces a version, measures that version against a standard, adjusts, and goes round again — without you having to read every attempt. You’ve stopped hand-cranking the work and started designing the machine that cranks it. The prompts don’t disappear; you’ve just stopped writing them one at a time.

Three things every loop needs

A loop that builds a page well rests on three decisions — and every one of them is a judgement call, which is exactly why this isn’t something a template store can sell you.

The objective is what “good” actually means for this site. Not “a website” — your website: this brand, this audience, this thing you’re trying to get a visitor to do. That’s the part your brief feeds — your logo, and the URLs and screenshots of work you admire — turned into a precise definition of done.

The metric is how the loop tells, on its own, whether a given pass came out better or worse — without us inspecting every pixel. Does this layout carry the brand or fight it? Is the hierarchy doing its job? Is it distinctive, or has it drifted toward the statistical average of every site ever made? A loop is only as good as its ability to mark its own work, and writing a metric the AI can actually apply is the real skill.

The boundary is how far the loop is allowed to run before it has to stop and come back to us. It’s where our risk tolerance lives — the line between “keep refining this on your own” and “this needs a human eye before it goes any further.”

Get those three right and the loop earns its keep: it asks, checks itself against the metric, adjusts, and keeps going — and what lands on our screen is already considered, not a first draft.

A loop that runs versus a loop that learns

Not every loop is worth the same. There’s a real difference between a loop that runs and a loop that learns.

A loop that merely runs is automation — it produces the same thing today it produced yesterday. Useful, but it doesn’t get better. A loop that learns knows whether each pass landed, and feeds that signal back into the next one. It compounds. Over two years and many hundreds of decisions, that’s the difference that built our system: every pass that worked, and every one that didn’t, has quietly sharpened what comes next. It’s the same discipline behind our vibe coding work, where we build custom software the same way — by refining the loop, not by hand-cranking the code.

Where we stay in the loop

If the loops do the work and judge the work, you might wonder what’s left for a human. The answer is: the part that actually matters.

Our job moved out of keystrokes and into judgement. We choose the objective, which forces a clear definition of what “good” looks like for you. We choose the metric, which is a decision about what’s worth optimising in the first place. And we draw the boundary, which is where decades of having seen where the AI is confidently wrong pays for itself. The loop tirelessly produces and scores; we decide what’s worth producing and how to score it. That’s the art of the prompt, turned into a repeatable system — and it’s why the same models anyone can sign up for produce something bespoke in our hands and generic in most.

The loop we hand to you

Here’s the part we like best: you get a loop of your own. The accuracy review built into Studio is exactly this idea, shrunk to a single button. You draft an article, Studio reads the whole thing and flags what’s worth double-checking, you fix it, and you re-run — write, check, correct, repeat — each pass a little tighter than the last. The machine does the tireless reading. You keep the judgement. Same philosophy that builds the site, handed over so you can run it yourself.

What it means for you

A bespoke website, built not by one lucky prompt but by loops that refine their own work against a standard we set for your business — and then a control panel that hands you a loop of your own to keep it accurate and fresh. The tireless part is the machine’s. The judgement stays human, on both sides of the handover. That’s the combination we think every business deserves, and it’s the conversation worth having.

Read next: Your website, in your hands: meet Studio — the control panel that turns this same loop into a single button you press yourself.