AI builds your webshop in a weekend. That is not the problem.
Building a webshop with AI now takes a weekend. You type what you want, the tool builds a site that works, and it is live before the weekend is over. The Swedish AI builder Lovable reached more than 100 million dollars in annual revenue within eight months of launch, faster than any software startup before it. Millions of people now build live sites without ever writing a line of code.
I am not saying you should not do that. I build AI systems myself. Milo, the agent that runs my whole operation, sits on a Mac Mini in my home and works 24 hours a day. I see every day what these tools can do, and it is impressive.
But there is a difference between a site that works in a demo and a shop that actually handles money. AI gives you the first eighty percent in a weekend. The last twenty percent is where your customers drop off, where your revenue leaks, and where most self-built webshops quietly fail. This piece is about that twenty percent: what it is, why AI skips it, and what it costs you if you ignore it.
The demo works. That says nothing.
The first eighty percent is the visible part. AI is good at it. A clean design, product pages, navigation, a shopping cart that looks right. On your screen it looks finished. You click through and think: this is done.
That feeling is wrong. What you test is the happy path: you, on your own laptop, with a product you entered yourself, no rush, no mistakes. The real world looks different. A customer on a slow connection. A payment that breaks halfway. Stock that is off. An address with strange text in it. An attacker checking whether your form leaks.
The eighty percent you see on your screen is exactly the part an AI can imitate from thousands of examples. The twenty percent you do not see is the part you only build by having lived through it. I spent six years building brands for e-commerce founders, and before that I ran my own brand. The pattern is the same every time: the shop does not fall over in the demo, it falls over on the first busy day.
Think of the ratio like a house. In a weekend the AI hands you a house that looks finished: walls, windows, a front door, a kitchen that looks right in the photo. What you do not see is whether the foundation holds, whether the pipes leak, whether the fuse box copes when everyone turns on the heating at once. You only notice when it goes wrong, and then you are standing in the basement. A webshop is no different. The visible part is done quickly. The invisible part decides whether you can live in it.
When it comes to building a webshop, the question is not whether it looks good. The question is whether it stays standing when something goes wrong. And something always goes wrong.
The twenty percent that hits customers: security
Start with the most invisible part. A webshop handles money, addresses and sometimes payment data. That makes it a target. And this is exactly where AI code fails most often.
In 2025 Veracode tested more than a hundred AI models on the code they produce. The result: 45% of the generated code failed security tests and introduced weak spots from the well-known OWASP Top 10. In 86% of the relevant cases the code did not protect against cross-site scripting, one of the most common ways to abuse a website. And the striking part: newer and larger models did not score better. Security stayed flat across years of model updates.
These are not abstract numbers. They are exactly the flaws you do not see in a demo and do feel the moment a real customer or a fraudster hits your checkout. A form that lets unchecked input through. A login that is too easy to bypass. A payment flow that leaks data.
The consequences are expensive. A broadly sourced analysis of the vibe-coding wave shows that thousands of companies that built fast with AI now need a rebuild, an expensive and time-consuming operation. A whole market has sprung up of agencies that repair AI-built apps. Building faster is not the same as building better.
For you as a founder this means something simple: the cheap route up front can become the expensive route the moment you have your first real customers.
The catch is in the timing. A security flaw does not show at the moment you ship the site. It waits. It waits until you have enough traffic to be interesting, until you process the first real orders, until someone comes along who sees where to look. That is exactly the moment you can least afford it, because by then you have customers who count on you. A shop that quietly falls over on day one is annoying. A shop that leaks customer data on the busiest day of the year is a different kind of problem.
Checkout: where your revenue quietly leaks away
Say your site is secure. Then there is still the checkout, and that is where most revenue disappears. Not because your product is too expensive, but because the last steps are off.
Baymard pooled fifty studies on cart abandonment. The result: on average 70% of shopping carts are abandoned. Part of that is normal, people are browsing. But among the customers who genuinely wanted to buy and dropped off, the reasons are concrete. 40% stop because of unexpected extra costs. 18% quit because they are forced to create an account. 17% find the process too long or too complicated. 9% miss a payment method they expect.
Look at that list. None of those reasons is about your product or your price. They are all about the build of your checkout. And they are exactly the details an AI skips, because they are not visible in a demo where you are the only customer.
For a Dutch webshop, the payment mix comes on top of that. iDEAL is essential here. Without iDEAL, or with a clumsy credit-card-only flow, you lose part of your buyers in the final second. Pay-later is gaining ground. The right combination depends on your audience, and that is not a choice a generic AI builder makes for you.
Baymard also states that most checkouts can drop a large share of their form fields. That is conversion gain without changing anything about speed or design. It is purely a matter of seeing what can go. I wrote about it earlier in why your webshop does not convert: the fault rarely sits in the design, and almost always in the assumptions underneath.
A webshop built for conversion starts at the checkout, not at the homepage. The homepage brings people in. The checkout decides whether they become customers.
Your brand: the one thing AI cannot copy
There is a reason this is the core of my work. An AI can build a webshop that works. What it cannot do is build a webshop that is yours.
Ask an AI for a clean, modern webshop and you get a clean, modern webshop. Exactly the same one a thousand others get who ask the same question. That is no accident, that is how these tools work: they average everything they have seen into a safe mean. The result is competent and completely interchangeable.
In a market where a customer decides in the first seconds whether to stay, interchangeable is the problem. I have seen enough webshops that were technically fine and still did not sell, because there was no reason at all to remember them. The product was good. The site worked. But the brand was empty.
Brand is not a layer you add on top later. It is the reason someone chooses your product over the identical product next to it that is a euro cheaper. That story, that positioning, that recognition: it does not sit in the average output of a model. It comes from choices made by someone who understands what makes your brand different.
This is also exactly where I am not anti-AI. I use AI every day. But I steer it with an idea. A tool without direction produces the average. A tool with a strong brand as input produces something that is yours. The difference is not in the tool, it is in what you put into it.
The invisible connections: stock, shipping, bookkeeping
Another part of the twenty percent that only counts once you are actually running: the connections. A webshop never stands alone. It talks to your stock system, your shipping partner, your bookkeeping, your email. In a demo that is all fake, with fixed test data. In production it is live, and that is where it breaks.
Your stock is at zero, but the site takes the order anyway. A shipping label is not created. A paid order does not land in your bookkeeping. These are not edge cases. This is the daily reality of a shop that sells. And they are exactly the things an AI-built site does not get right on its own, because they depend on your specific situation.
I see this on almost every project. The client comes with a site that is "basically finished" and only needs "the finishing touches." Those touches are invariably the integrations, the error handling and the edge cases. On Drivv and Maoo the real work was not in how the shop looked, but in whether it kept holding up when a hundred orders came in at once.
This is the unglamorous part of webshop development. It does not produce a nice screenshot. But it is the difference between a site that sells and a site you have to fix by hand every Monday morning.
When you should just use AI
For a lot of things, building with AI is simply the right choice. A landing page to test an idea. A simple one-product site for a temporary campaign. A prototype to show an investor or yourself how something could work. In those cases speed is the point and the risks are low. Building with AI is the right call, and you would be crazy to sink weeks of work into it.
The line is at the moment real money runs through it and real customers land on your site. From that point the things you do not see weigh heavier than the things you do. A prototype that falls over costs you an afternoon. A webshop that falls over costs you customers who do not come back, and a reputation you do not fix with a restart.
It is the same choice I make with photography. For quick testing on a marketplace, AI wins. For the image where someone has to be able to feel your product, I choose a real shoot. Not out of principle, but because the situation decides. With webshops it is the same: use AI where speed counts, and build production-grade where it hits your customers.
What production-grade really means
"Production-grade" sounds like jargon, so let me make it concrete. It is not a seal of approval, it is a list of things that are right when no one is looking.
Payments are processed securely and sensitive data does not leak. The checkout carries the right Dutch payment methods and as few hurdles as possible. The site does not fall over at a peak. Stock and shipping match, and an error is caught instead of breaking your whole shop. Your webshop stays findable, so a migration or rebuild does not throw away your existing positions in Google. And your brand is consistent everywhere, from the first impression to the confirmation email.
That is the twenty percent. It is not exciting and it does not produce a nice demo. But it is the part where a shop makes money or loses money.
An AI gives you a starting point that looks impressive. That is real value and I would never say otherwise. But treat it for what it is: the start, not the finish. Anyone who wants more money to run through it than they are willing to lose builds the last twenty percent with the same care as the first eighty. For a Shopify webshop the price is clear, and that amount is nothing next to the cost of building twice.
Want to see where your webshop leaves the last twenty percent on the table? Book a call, and together we will look at your checkout, your connections and your brand, and tell you honestly what can and cannot be done with AI.
