For businesses
What decides the cost of building a site or app with AI
No price table: what AI really made cheaper in building a site or app, what it did not, and what to ask anyone who quotes you. From a real project.
Search for "how much does it cost to build a website" or "how much does an app cost" and you get a lot of price tables. Their ranges sit far apart, and almost none of them explain what changed once AI started writing a large part of the code. I won't give you a number here. I will explain what actually moves the price, so you know what to ask anyone who quotes you, me included.
Why is there no price here?
Because I don't price anything before there is something to see. First I build a working first version, and only after you've seen it and decided to go on do we talk about next steps and payment. The first version costs nothing. A price set before anyone has opened a single screen is mostly a guess, and it usually carries a large safety margin to cover that guess.
What did AI actually make cheaper?
Typing the code. A screen, a form, a database table, the wiring between two parts: what used to take days now takes hours. When I built a room booking system for the Shiluv clinic, the first commit went in on 17 September, and the same day one booking already went all the way through, from the clinic's site to the staff's screen. Twelve days later the system was live.
So "building the screens" is no longer the big line on the bill. Anyone who prices a site or an app by the number of screens, as if each one still took a week, is pricing the past.
What does AI not make cheaper?
The decisions. What exactly the system does, for whom, and what happens at the edges: a last-minute cancellation, two people booking the same room in the same minute, a client who wants the same slot every week. That has to be worked out with someone who knows the work, and AI can't ask those questions for you. At Shiluv this was built in rounds: the team tried a version, wrote to me about what didn't fit, and I fixed it. That's how things like the latest hour a room can be rented were settled. I wrote more about this in Domain Driven Vibe Building.
AI doesn't make the checking cheaper either. Code written fast needs someone to check it. A 2025 Veracode study tested more than a hundred language models on 80 coding tasks and found that in about 45 percent of cases the code they wrote introduced a security flaw. Shiluv has 54 test files today, and they are part of the work, not an extra.
How many screens and features does a first version need?
Fewer than you think. The biggest factor in the price is scope, and the right scope for a first version is the one everyday action the system exists for. For a business website that's usually "people understand what I do and get in touch". For an app it's the thing the user does when they open it at the bus stop. Everything else can wait for the next version, and then you add it based on what you learned from real people, not on what seemed important up front.
Do sign-in, payments and integrations add cost?
Yes, and this is where it pays to ask the most questions. Every connection to a system you don't control comes with someone else's rules. A payment provider requires an onboarding process and checks. WhatsApp Business requires a Meta business account and message templates that Meta approves. Google Calendar needs permissions. At Shiluv every room has its own Google calendar connected to the system, and that connection took more thought than any screen on the site.
User sign-in changes the picture too. A brochure site needs no login. A system with staff, clients and different permissions for each one does. At Shiluv the login is a code sent by email, with no passwords, but behind it sits a decision about who sees what.
What does it cost when there's data about real people?
This is the line that's easiest to forget. Once a system stores names, phone numbers or bookings of real people, you need backups, you need a separate test environment so a change doesn't break something for clients, and you need the right permissions in the database. Shiluv has a nightly backup to the clinic's own storage, and every change is tested first on dummy data. A brochure site with no forms skips most of this. A booking system doesn't.
What does it cost to keep it running after launch?
There are two kinds of running costs. The first is infrastructure: the domain, hosting and the database. For a simple site they're usually small, and sometimes covered by hosting providers' free plans. The second appears only if the product uses AI: every question the system sends to a language model costs money, and the bill grows with use. It's worth knowing up front who pays for it and whose account it's on.
My answer is always the same: the system runs on your accounts. At Shiluv the servers, the database and the calendars are registered to the clinic. I built it, they own it, and the running bills go to them directly, not through me.
Fixed price or hourly?
Both try to solve the same problem: nobody knows up front how much work there is. A fixed price moves the risk to the builder, who prices it in. An hourly rate moves it to you. I'd rather shrink the risk itself: a working first version, and then we agree on the scope for what comes next, when we can already see what's there and what's missing. That way the quote rests on a product, not on a document.
How do you get a real estimate?
Write a few lines about what you need, in your own words. No spec and no technical language. I build a first version, and once you've seen it you'll know exactly what you're buying before you pay. And if you're talking to a few people, it's worth going through what to check before you hire someone who builds with AI first.