AI-Built Websites: What Small Business Owners Should Check

AI tools now write code so fluently that a website or small app can be assembled in an evening without anyone reading a single line. For a business owner, that sounds like savings: why hire a studio when you can describe the idea to a chatbot? We use AI tools every day and see both sides — where they genuinely speed things up and where someone later has to clean up the mess. Here is what to check before you hand an AI a project your customers and revenue depend on.

Where the "just ask the AI" trend came from

In February 2025, researcher Andrej Karpathy coined the term "vibe coding": you accept whatever the AI assistant suggests and stop thinking about the code altogether. The phrase spread fast and became something of a word of the year. It was, however, a tongue-in-cheek description of weekend experiments and throwaway projects, not a method for building products real customers rely on.

Today plenty of contractors and freelancers work exactly this way: the AI writes, a person clicks "accept". That is neither good nor bad in itself. What matters is knowing which jobs it suits and which it does not.

Where AI-built projects make sense

Some tasks put speed first and carry little cost if something goes wrong. For these, a quick AI-assisted build is a sensible choice:

• a prototype to show an idea to partners or an investor;
• a simple landing page to test demand, with no accounts or payments;
• an internal tool used by two or three employees;
• a hypothesis check, such as "will anyone actually request this service?"

If the idea fails, you lose little. If it works, you find out quickly and cheaply.

A fast AI build is fine as long as you could throw the project away without regret. The moment customers, payments or personal data appear, the rules change.

What breaks once customers and money arrive

An AI will happily announce that "the app is ready". It will not ask whether your customer database is locked down or whether one user can see another user's orders. Those are questions an experienced developer asks, not a chatbot.

A few well-known cases show how this plays out.

A service shut down within days

The founder of Enrichlead publicly boasted of building the product entirely with an AI assistant, without writing a line of code himself. Soon users were bypassing the paid subscription, basic security holes were found, and the service had to close. The problem was not the AI itself — it was that the person signing off on the result had no way to verify it.

A deleted database

In July 2025, SaaStr founder Jason Lemkin described how an AI agent deleted a live production database even though it had been explicitly told not to change anything — and then misreported what had happened. The lesson is simple: a written instruction to the AI is not protection. The only real protection is making sure the tool has no access to critical data in the first place.

Parts that do not exist

Modern websites are assembled from ready-made components, much like furniture from standard fittings. AI sometimes confidently recommends a component that does not exist. According to a 2025 study cited in this Habr review, roughly 20% of such suggested components were, on average, invented. Attackers exploit this by registering a malicious component under the same name and waiting for someone to install it.

The most expensive part is maintenance

Launch is only the beginning. A website needs ongoing work: a new section, a revised order form, a new delivery provider. This is where a project nobody has ever read starts to cost real money.

The AI fixes one thing and quietly breaks another, because it does not remember why anything was built the way it was. The person who "just described the idea" cannot point to where the problem might be. Every "please fix it" becomes a lottery: sometimes it works, sometimes it adds hundreds more lines that nobody will look at either.

For the business, it looks like this:

• small changes turn out surprisingly expensive and slow;
• a new developer opens the project and suggests rebuilding it from scratch;
• bugs pop up in random places after every update.

"We'll rewrite it later" almost never happens: once the site has customers and orders, a rewrite is a major project of its own. We covered what drives development and maintenance costs in our article on website pricing in 2026.

"AI makes us 10x faster" — does it?

Many contractors justify their timelines and prices by pointing to AI. That is partly true, with a caveat.

In a 2025 METR study, experienced developers worked on tasks in their own projects with and without AI tools. With AI, the work took about 19% longer — yet participants were convinced they had been faster. The tools have improved considerably since then, so "AI slows everyone down" is not the right conclusion. But feeling fast and being fast are clearly different things.

There is a simpler explanation too. Writing code is only one part of building a website. There is also scoping, design, approvals, testing, launch, server setup, security and support. AI dramatically sped up only the first part. Speed up one stage tenfold while leaving the rest unchanged, and the project as a whole gets only modestly faster.

On top of that, code now appears faster than anyone can read it carefully. If a contractor saves time on review, you get something fast, but not necessarily right.

Be wary of reports like "AI wrote 90% of the code" or "we made this many AI requests". Those measure activity, not results. What matters to you is whether the site works, brings in leads, and what it costs to grow.

A hidden trap: licenses

If you are ordering not just a website but a product with AI inside — say, a chatbot that answers customers using your documents and price lists — there is another risk that rarely comes up.

Such systems are usually built from free, ready-made components. But "free" does not always mean "use however you like". Some popular components come with terms that require you to open the source code of your entire product if people access it over the internet. Others forbid commercial use outright. Some oblige you to display a visible notice on your site naming the technology behind the product.

While the system stays inside the company, few people care. Problems start when you want to sell the product, raise investment, or a large client sends a questionnaire asking you to list all software in use and its licenses. That is when you discover part of the project needs urgent rework.

Questions to ask your developer

You do not need to understand code to judge how reliable a contractor is. Ask a few questions and notice how confident and specific the answers are:

1. Who reads and reviews the code the AI writes? Does that person have a name and real experience?
2. How is customer data protected? Can one user see another user's data?
3. Who has access to the live site and database, and could an automated tool delete anything there?
4. Will you set up backups, and how quickly can you restore the site after a failure?
5. Could another developer understand the project if we decide to switch contractors?
6. Which ready-made components are used, and do their licenses allow commercial use?
7. How will you measure results — not the volume of work, but leads, sales, site speed?

If half the answers boil down to "it's all fine, the AI handled it", take that as a warning sign. For more criteria, see our guide on how to choose a web studio.

Why work with us

We are not against AI. We use it wherever it saves our clients time and money. But responsibility for the result always sits with a person, not a program.

• A developer reviews every line. AI speeds up routine work, but the code that reaches your site is read and understood by people.
• Security is built in from day one. Access control, customer data protection and backups are part of the project, not a patch after the first leak.
• Your project stays maintainable. Hand-written, readable code means we — or any other team — can keep developing it.
• We check licenses. If your project includes AI features, we choose components you can legally use in a commercial product.
• We measure results, not activity. We agree on goals — leads, sales, search rankings — and report against them.
• We give honest advice. If a quick prototype is all your task needs, we will say so and not sell you more than necessary.

The bottom line

AI is a great tool for fast experiments and a poor substitute for an accountable developer. While a project is disposable, save money boldly. When customers, revenue and reputation are at stake, you need someone who understands exactly what is running on your site.

If you want to bring AI into your business without unnecessary risk, see how we approach AI integration. To discuss your project, leave a request through the contact form on our home page — we will get back to you and suggest the most sensible place to start.