AI made building cheap. The thinking is still the expensive part.

In January 2025, right at the peak of the AI hype, I started a design agency. Friends asked if I had missed the news. Anyone can generate an app now, so why would a company pay people to design one?

Almost two years and a lot of projects later, I have a clear answer, and it comes from the work rather than from theory.

The calls we started getting

A pattern showed up in our inbox in the first year. A founder or a product lead would get in touch with a product that already existed. It had been built quickly, often with AI tools, and it looked fine. Clean screens, sensible colours, a login that worked.

Then the same sentence, in different words: nobody is using it.

When we dug in, the problem was almost never the interface. The product answered a question nobody was asking, or it answered the right question in a way that did not fit how people actually work. The team had skipped straight from idea to screens, because screens had become cheap enough to skip to.

What got cheap, and what did not

AI has made a lot of our work faster, and we use it every day. Generating a layout, writing a first version of a component, producing variations of an illustration: tasks that took days now take hours. We pass that saving on to clients, and it is a big part of why a small senior team can deliver what used to need a department.

What AI has not done is tell anyone what to build. It will happily produce a beautiful onboarding flow for a product with the wrong onboarding. It does not sit in a call with a nurse, an insurance partner or a finance team and notice the hesitation before they answer a question.

Clients who build it themselves end up paying for the thinking they skipped. Usually later, and usually more.

Where the value of design actually sits

For years, the visible part of design was the output. The Figma file, the polished screens, the brand guidelines. It was easy to believe that was the job, because it was what clients could see.

The real job was always upstream. Understanding who the product is for, what they are trying to get done, what stops them today and what would make them come back. Turning that understanding into a structure: which features matter now, which ones can wait, how the pieces connect. Only then does it make sense to make it beautiful.

When outputs became cheap, that upstream work became the whole difference between a product that works and one that only looks like it does.

How we work because of it

On every new product we start with the problem space, even when the client arrives with a finished brief. That usually means a short phase of conversations, user flows and a clear definition of what the first release has to prove. On WTE Companion that groundwork shaped a chat-based app, two operational portals and the website around one question: what does someone moving to Germany need to get done first?

The result is now live with ten partners, including insurers, a bank and a mobile provider, processing applications through the product. None of that came from generating screens faster. It came from deciding what the screens needed to do.

What this means if you are building something

• Use AI to go faster, without treating it as a reason to skip the thinking. Speed only helps once you are pointed in the right direction.
• Write down what your first release has to prove before anyone designs or builds it. If you cannot, that is the work to do first.
• Talk to the people who will use the product, even five of them. You will learn more in those five conversations than from fifty generated layouts.
• Judge a design partner by the questions they ask in the first call, not by how quickly they show you screens.

Good design has always been about solving the right problem. AI just made it impossible to hide behind the polish.