Building your website with AI: the tool builds it, but you do the work

You open Lovable, type one line: "build a website for my removals company, with a quote form." Half a minute later there is a site. No developer, no agency, no waiting. It looks sharp.

That is how a growing number of business owners build their website: not through an agency, but with an AI tool like Lovable, Bolt, v0 or Claude. And some go further and let an AI agent like OpenClaw or Hermes run their marketing. You describe what you want, the machine does the rest. It is called vibe coding, and for a quick prototype it is brilliant.

And then you sit there looking at it. It looks fine. But is it any good? That question is where it goes wrong, and it never really goes away.

The idea: you describe in plain language what you want, and an AI builds it, without you looking at the code. It is called vibe coding, a term from AI researcher Andrej Karpathy (2025). For you as a business owner it looks like this: you open Lovable or Claude, type what kind of site you want, and one appears. Not right? You ask for a change, over and over, until it seems to do what you want.

The tools you hear about

There are two flavours. For your website, the builders like Lovable, Bolt, v0 and Replit, or an assistant like Claude or ChatGPT that writes the code for you. For your marketing, the AI agents like OpenClaw and Hermes: you let an AI carry out tasks itself, from writing copy to adjusting campaigns. The promise is the same for both: describe it, and the machine does the rest.

There is a site. It looks sharp. And then comes the question that sticks: does this actually work? Not "does the button click", but: does this bring in customers? Is the right thing on there? Is it any good?

And that is where you get stuck, and it is not for lack of intelligence. You are not a developer, so you cannot see whether the code is sound. You are not a marketeer, so you cannot see whether that headline does anything. You see the outside, and the outside always looks good. That is precisely what an AI is best at: making something that looks right.

You do not know what you do not know

And that is the nasty part. A site can be quietly mediocre for months without you noticing. No error, no alarm. Just: slightly too few requests, and you assuming it is the market. You cannot judge what you cannot see, so you stay parked at "looks good, I think?"

Why do you have that feeling? Because the thing is guessing. Building it can do. Knowing what belongs on it, it cannot, and that is the hard half.

Because you know that from data that sits somewhere else entirely: what your customers actually search for, which page brings in requests and which does not, what a click costs you in Google Ads, what your competitors are doing, which text worked last quarter. Lovable, Claude or an AI agent knows none of it. They do not know your business, they know the average of the internet. So the average of the internet is what you get back.

Wiring that data up yourself? That is rebuilding the engine

You could solve it by connecting all those sources to the AI yourself. That is exactly what agent frameworks like OpenClaw and Hermes promise. Only: then someone has to hook them up to every data source, give them a memory of what was already tried, teach them to judge whether something is an improvement, deploy it and keep watching it. Months of specialist work, none of it about your business. You would be rebuilding an engine that already exists, just to reach the starting line.

And then the part nobody budgets for: the time. Not because it is heavy work, but precisely because it does not feel like work.

Add it up and that half minute of magic has become a few weekends. Spent on something you still cannot tell works.

This is not an "AI is bad" story, quite the opposite. Even Karpathy, who coined the term, later said he meant vibe coding mainly for quick weekend projects. As tooling it is fantastic, as long as you use it where it is strong:

• Trying an idea quickly. Want to see whether a concept works before you invest in it? In an hour there is something to react to. Perfect.
• A throwaway version or an internal helper. A landing page for one campaign, a simple internal tool, something with no customer data or brand attached. Fine to vibe-code.
• Learning and sketching. It is an excellent way to grasp what is possible, and to make an idea tangible for a conversation.
• With an expert at the wheel. In the hands of someone who can read the code, knows what to feed it and can judge the result, it is a huge accelerator. The tool is not the problem then, being unable to judge the output is.

The line is simple: use it for things you can afford to be wrong about. The moment customers, data or money run over it, you want someone who can tell whether it is right.

We run it as a supervised system. The machine is already built and runs on all of your data in one place, so you connect nothing. But the real difference is not the technology.

The difference is that you no longer have to answer that question, does this actually work, yourself. A specialist approves every change and can see whether it converts. If something goes wrong, that is our problem, not yours. You message what you want, we do it, we check it, and we are the ones you hold to it.

And that is not theory. Our own site is built and improved by that same system every day, with a person on it. For clients it produces numbers you can simply go and check:

None of them had to wonder whether it was working. It was there in the numbers. See all results