I recently attended the PUSH UX conference in Munich, and one slide from Kris Hedstrom, Head of UX/UI Design at On, has stuck with me ever since:
“We can build almost anything. Should we?”
It sums up a shift I see in my own daily work. Building has become cheap, and the technical process is no longer the bottleneck. So the job of designers and developers shifts from making to deciding: what belongs on a page, what gets left out, and how we shape the flood of AI output so people can still make sense of it.
Every element has to earn its place
More features don’t automatically mean a better product. An overloaded UI can achieve the opposite of what was intended: users get overwhelmed and can no longer find what they came for.
Sometimes the best design decision is to leave something out. A calm, focused page protects a product’s perceived value. Premium brands know this well: the less there is on the page, the more valuable the product feels.
Putting numbers on it
Kris showed how they measure what is actually earning its place on a page. On a product detail page, for example, they track every element: how often people click on it and how much those clicks are worth in the end. “Value” here basically means: how likely is it that someone who clicks on this element ends up buying?
They plot both on one chart and every element lands in one of three groups:
- Engagement drivers: high click-through, lower value per click. People use them a lot to explore, like color thumbnails or the size chart.
- Value drivers: fewer clicks, but a higher expected value per click. In the example, people who engage with “How to wear it” generate noticeably more value per click, even though few of them click it.
- Neither: low on both. These are the candidates to simplify, merge or remove.
The chart below is a simplified version of the example from the talk.
Personally, I found this approach really exciting, and I want to try it in one of my own projects. The prerequisite, of course, is that you already have solid tracking in place. Once you track every interactive element and can link it to purchases, you can build a chart like this yourself. One caveat: it shows correlation, not cause. People who open “How to wear it” may simply be closer to buying already. Still, it’s a great tool for conversations with stakeholders. It shows which features can be removed, and it makes room to try new things without cluttering the page.
Designers are moving beyond Figma
Apart from the talks, I talked to a lot of people at the conference. One thing came up again and again: more and more designers are moving away from Figma. You read about this all the time on LinkedIn, and to be honest, I always thought it was just the LinkedIn people who work like this. But it was crazy to see how many actually do it in real projects. Designers who always worked in Figma now pull the codebase locally for the first time and build their prototypes directly with Claude.
They use these prototypes for user interviews and then hand the branch over to the developers, who finish the feature. One designer told me he now opens pull requests himself for small design changes. Just like developers no longer need to know every CSS property by heart, designers no longer need to know every trick in Figma. It’s the same shift again: building a prototype is no longer the hard part. The hard part is deciding what is worth building at all.
Too much output is a UX problem
As a developer, I run into a similar problem every day: LLMs produce more output than we can take in. Think of the pull request with 120 changed files that a local agent generated (LGTM). Reviewing these huge PRs annoys us developers all the time, but I never thought of it as a UX problem. It is one, though.
The question is the same as on the product page: what actually earns its place? We need well-established UX practices to shape that output:
- What is relevant in this diff, and what can be collapsed?
- Which changes affect which part of the system?
- How should an agent’s output be structured so a person can review it?
And it’s the same with the things we build for our customers. If we build a chatbot that answers user questions, our users get just as annoyed by a giant wall of text as we do. So we have to think about how to present the answer in an attractive way. Take a shopping assistant: instead of a long text, it should show a few product cards, built from components of our existing design system. That’s what dynamic, generative UI is about. The information becomes tangible again, and the user stays in control.
Less making, more deciding
Measuring elements on a product page, designers prototyping directly in code, agents writing huge PRs: it all comes back to the same point. We can build almost anything now, and it’s getting easier every month. So the real work for UX, and honestly for us developers too, is deciding what deserves a place. And if we can back those decisions with data, even better.

