Anyone can create a website or an app with AI today: that is where the work begins
With AI, I can create a website, an app or a whole SaaS in an afternoon. What AI won’t tell me is whether anyone wants it and whether people understand it. Why that is the most important question today, and how I answer it.
Martin ProkopFounder of Tack··6 min read
Short answer
AI has cut the way from an idea to a working product from months to hours. The bottleneck has moved with it: it is no longer creating, it is validating.
AI creates a website or an app exactly to your brief. Whether the brief was right, whether anyone wants the result and whether people can use it, it won’t tell you. Only people can.
So I put every version in front of a few people before I spend more time on it, and after launch I ask those who actually came. That is exactly what I created Tack for.
On this page
Creating a website or an app is no longer the hard part
Only a few years ago, a product idea meant months of work. A brief, a design, development, testing, and only then the first user. Today I open Lovable, v0 or Claude Code and by the evening I have something to click through. A working website, an app with sign-in, a whole SaaS with payments.
That is great news. Far more people can try their idea today: a designer without a developer, a product manager without a budget, a small agency for a client who wants to see results within a week. But as creating got cheaper, what is expensive changed.
What AI won’t do for you
AI is an excellent executor. It creates a website or an app exactly to the brief, and fast. There are three things it won’t do for you:
- Interest. AI creates an app, not the people who want it. Whether the problem you are solving really hurts anyone, the code won’t tell you.
- The right design. AI optimises for the brief, not for needs. Brief the wrong thing and you get it fast and well made.
- Usability. A generated interface looks finished. That is the trap: a finished look tempts you to skip the question of whether people find their way around it.
A prototype that looks like a finished product is more dangerous than a sketch on paper. Nobody mistakes a sketch for an answer. A polished prototype, they do.
The bottleneck has moved
When creating a website or an app takes an afternoon, the costliest mistake is no longer bad code. It is the week, the month, the campaign and the energy put into something nobody wants, or something people can’t use. How fast we create has outrun how fast we learn about what we created.
„AI helps us create a website or an app today. It won’t create people’s interest or a fully working product for us. That still takes human interaction, and that is a good thing: we don’t make products for AI, we make them for people.“
The advantage doesn’t go to whoever creates the most websites and apps, but to whoever learns fastest which of them make sense to people, and drops the rest before they cost more than an afternoon.
How I do it
The process I use for my own products and for clients has five steps:
- I create the smallest version of the website or app that answers one question. Not the whole product, just the part I am unsure about.
- I give it to five people with one specific task. Five people in one round uncover most usability problems (Nielsen Norman Group). Several rounds of five beat one big one.
- I measure, not just listen. Did they finish the task? How long did it take? How hard did it feel? People say one thing and do another.
- I fix it and run it again. With AI a fix takes minutes, so the next round can run the same day.
- After launch, I ask those who actually came. One question at the right moment tells you more than a long survey sent by email.
Why I created Tack
User research tools are excellent, but most of them expect a research team, designs in Figma and a yearly plan. For someone who created a prototype this afternoon and wants to know tomorrow whether it works, that is needlessly complex and expensive.
So I made Tack, which I develop with the help of AI myself, as simple as possible. One script for a prototype and a live website alike, templates with standard questions, results by the next day, and you pay per test, not per month. So that validating a website or an app is as easy as creating it.
FAQ
Why test a product that AI created?
Because AI creates exactly what you ask for, but it won’t tell you whether people want it or can use it. A generated product also looks finished, which tempts you to skip validation. A short test with a few people shows where they get stuck before you put more time into the product.
How many people do I need to test a prototype?
To find usability problems, about five people per round is enough. Several smaller rounds work better: test, fix and test again. For numbers such as the share of people who finish a task, you need a larger sample, roughly thirty people or more.
How do I test a prototype made in Lovable, v0 or Claude Code?
The prototype has to run at an address people can open in a browser. Add the Tack script to the page, pick a test template, set when the task counts as done and send the link to five people. You see the results by the next day.
Martin ProkopHead of Design at Czech Television, founder of Tack
I am Head of Design at Czech Television and the founder of Tack, which I develop with the help of AI. Before that I led design and research at Livesport, on a product used by more than 100 million people a month, and worked for Seznam.cz, Home Credit and ČSOB. In 2020 I founded the travel platform Retkee, which I still run.