Anyone can build a prototype with Claude now. So why hire a designer?

Because a prototype is a stack of decisions: what problem it solves, who it's for, which patterns people already know, how far to take it and whether it fits the rest of the product. Claude builds fast. It doesn't make those decisions for you.

And together they decide one thing: whether people trust the product.

People judge a product before they read a word. In one study, they judged how a web page looks in 50 milliseconds. In Stanford's study of 2,684 people, the look of a site came up in 46% of what they said about its credibility, more than anything else.

A product that looks right is seen as easier to use, and people forgive it small problems. One that looks off loses money: 19% of US shoppers who dropped an order at checkout did it because they didn't trust the site with their card.

Banks show it best. Trust used to come from the banker's suit. Now it comes from the app. In a study of mobile banking, better-looking apps earned more trust, and people were more ready to use them.

Robinhood started in 2013 with no customers and no banking history. Its app won an Apple Design Award in 2015. Today it has 28.6 million funded customers. Bank of America has 4.2 million investment accounts and Chase 1.7 million, and they are the two largest banks in the US. Price doesn't explain the gap: Chase and Bank of America's Merrill charge $0 for online stock trades too.

Even a crisis didn't break it. In January 2021 Robinhood stopped people from buying GameStop, and the anger was everywhere. By the end of that year it had 22.7 million funded accounts, up from 12.5 million.

People pay more to companies they trust and leave the ones that let them down. Nearly two-thirds of business buyers who trust a company will pay it more. For 81% of people, trusting a brand decides whether they buy from it. And for 32% of people, one bad experience is enough to leave a brand they love.

Companies that take design seriously grow faster. In McKinsey's five-year study of 300 companies, the top quarter by design grew revenue 32 percentage points more than others in their industries.

People describe a solution, not their problem. Ask them what they want, and they'll ask for a faster horse that eats less, as the line credited to Henry Ford goes. Nobody would have asked for a car. The real job was getting from A to B faster.

Or, as an old marketing line goes, people don't want a quarter-inch drill. They want a quarter-inch hole, to hang a shelf.

People trust a product that does the job they came for. Finding that job first is also cheaper than rebuilding a feature nobody uses, and most features are exactly that. In Pendo's study, 80% of features in the average software product were rarely or never used. In a Forrester study, testing with users before launch saved a large company $2.5 million in developer rework over three years.

In Hochuapp, my own app, the problem wasn't a missing number. An exchange already shows one total, and a red total only says stop. What people couldn't see was what each purchase did. So Hochuapp shows every purchase on its own: which ones already carry you and which are still waiting. My brother sees that his 2023 bitcoin covers the drops of his later ones, so the news about bitcoin falling doesn't throw him.

A product doesn't live on its own. Every day people use dozens of other apps and expect yours to work the same way. That's Jakob's law. Anything unfamiliar feels like a wall. Anything familiar gives people a sense of control: they know what happens when they tap.

That's why Apple and Google publish detailed guidelines for iOS and Android: when apps behave alike, people open a new one and already know how to use it. IBM spent three months moving its checkout to the standard patterns of its Carbon design system, and conversion went up 5%, with only minimal testing.

Games are the exception. In a game, surprise is the fun. In a tool people work in, it slows them down.

At Punto Pago, a fintech launching credit cards in Panama, only about half of people trusted banks. The card had to feel safe and normal. Most of their audience was on Android, where the old site was heavy and slow, so I designed a new landing page that opens easily there. With a campaign around it, conversion to the waitlist went up 55%, before people could even get the card.

Claude answers with what it has seen most often. A screen library like Mobbin keeps strong and weak solutions side by side, and without knowing which is which, Claude goes for the middle. On a real product, with its own limits, the middle drops below average.

The problem isn't access to information. It's too much of it. Medical textbooks are open to everyone, and that doesn't make us doctors. Picking the right pattern for one task out of all of them takes years of practice.

The right pattern is the one people get at first sight, and that's what makes them feel safe. At Selzy, the agents can look through Mobbin, but the designer picks what fits our users.

People trust a product that behaves the same way everywhere: one task, one way of doing it, on every screen. If one thing works differently on two screens, people stop trusting both. That's what a design system is for, and it pays back in money.

In a report by Figma and the Design Executive Council, a global social media brand saved $16 million a year with its design system only 60% adopted, and $67,000 on a single feature: a rollout that usually took three months took four days. In a 2026 study, developers built the same interfaces 1.6 to 2.5 times faster with a design system than by hand. With AI working from the system, they got 96% of each task done, against 68% by hand. When Coinbase had its AI agent build from the kit's own components, build time and AI costs both went down by about 22%.

At Selzy, prototypes had drifted from prod, and a developer once got three versions of one screen: Figma, the prototype and prod. So I made it a rule that a prototype has to match prod. I built one layout template from the product's own components and wrote a script that measures the live product: it caught about 30 small differences that had slipped through. Interface text comes from the product's real translations, so every language shows up in the prototype, and long strings get fixed there, not in prod.

At Selzy, an email marketing platform with more than 50,000 customers, I lead a team of four designers. Since May we build prototypes in code, on the product's real UI kit. Claude does the building. We keep the decisions: what problem a task really solves, which pattern our users already know, and whether the screen matches prod in every language.

That's what a design engineer is to me. The agents build. I answer for every decision, and for the trust they add up to.