An AI website builder will now produce something publishable in an afternoon, and it will look fine. Pages, copy, a colour scheme that does not embarrass anyone, a contact form that works.
That is real progress, and it has eaten a category of work that used to be billable. For many simple brochure sites, the build is no longer the constraint.
The interesting question is what did not get solved.
The site was never the hard part
The work that survives is not “make a nicer page”. It is everything that happens after somebody fills in the form.
Where does that enquiry go. Does it reach the person who handles it, in the system they already work in, with the fields they need. What happens if it arrives at 11pm on a public holiday. Who finds out if it silently fails.
None of that is web design. All of it is the actual project.
Generators produce the surface of a business. The connective tissue underneath is where the hours go, and it is specific to your company in a way no model can infer from a prompt.
What can an AI website builder not do?
It cannot obtain authority over your business accounts, decide rules nobody has written down, or carry responsibility when something breaks at 2am. It builds the page. Connecting that page to how your business actually runs is a different job.
Anything that has to talk to something else
This is the largest category, and the one most consistently underestimated.
Every integration is a negotiation with somebody else’s system. Their authentication, their rate limits, their idea of what a valid date looks like, their habit of renaming a field in a minor release. A generator can produce a form. Deciding how to retry a failed webhook, or what to do when an API returns success but the record never appears, is judgement about your business rather than about markup.
Messaging is a good example because it looks trivial and is not. Sending an order update over WhatsApp means working inside Meta’s template rules and pricing model, where the WhatsApp Business Platform pricing documentation governs which message categories cost money and when. Get the category wrong and you either annoy customers or pay for messages you did not need to send.
I built Uwrapz around exactly this problem. A car wrapping business where vehicles sit in the workshop for days, and the owner spent his time answering the same question by phone. The website was the smallest component.
What mattered was stage tracking and a public status link for each job. Then notifications that fire when work moves, which meant deciding what counts as a stage change and what a customer should actually be told. By my own project records it was a solo build that reached production in about two months, and very little of that time went into pages.
No prompt describes that system, because until you build it nobody has articulated the rules.
Rules that only exist inside somebody’s head
Ask a business how its pricing works and you will get a clean answer. Then you build it, and the exceptions arrive.
Trade customers get different rates. There is a discount that applies only when two specific items are bought together. Orders above a threshold need approval, except from the two people who do not need approval. The delivery fee waiver has been applied inconsistently for three years and nobody wants to be the one who changes it.
This is not documented anywhere. It emerges through questions, and the questions are the job.
A generator has no route to it. It can only build what it is told, and the defining feature of these rules is that nobody can tell you all of them upfront. Extracting them is a conversation with the people who run the business, held over weeks, usually while they discover their own inconsistencies.
That is consulting wearing a developer’s job title, and it is the least automatable part of the work.
Where being wrong gets expensive
Some parts of a website carry consequences that make “good enough” the wrong standard.
Payments are the obvious one. Depending on where your business and your customers are based, card flows may need to satisfy requirements such as strong customer authentication, and they have to handle declines and partial refunds sensibly. How you integrate the checkout also determines which PCI self-assessment applies to you, as Stripe’s security guide sets out, and that is decided by an architecture choice made early.
Structured data is a quieter version of the same principle. Google’s structured data documentation is explicit that markup must reflect what is visible on the page, and generated schema that drifts from the visible content can make the page ineligible for rich results.
Then there is anything touching personal data, where the cost of a mistake is not a bug report.
These are not areas where you want output that is statistically plausible. You want output that somebody is accountable for.
Who answers when it breaks
The last category is not a capability at all. It is ownership.
Software fails at inconvenient times, and someone has to have context, access, and a reason to care. A generated site has no such person attached to it. When a payment gateway changes an endpoint or an update takes down the checkout, the tool that built the site is not going to notice.
That is the real distinction, and it is not “AI versus human”. It is whether anyone is responsible for the thing continuing to work.
For a simple site, nobody needs to be, and using a builder is the correct call. The honest comparison of where AI builders pay off and where they bite covers that decision properly.
Use the generator for the website. Hire a person for the parts your business actually runs on.
Common questions
Can AI replace a web developer?
For static, self-contained websites, largely yes. Where a project is dominated by integrations, undocumented business logic, payment compliance or ongoing accountability, the remaining work is defining requirements and connecting systems, and that is where the effort concentrates.
Is it cheaper to build a website with AI than hire a freelancer?
For a brochure site, yes, and often by a wide margin on the initial build. The comparison narrows once the site has to connect to other systems, because that work is priced by complexity rather than page count and does not shrink when the pages get generated.
What should I use an AI website builder for?
Validating an idea, launching a campaign page quickly, or running a small service business site that does not integrate with anything. Those are genuine wins, not compromises.
If you already have a site and the problem is what happens after the form gets submitted, that is a solvable conversation. Send me the specifics on WhatsApp.