Bart Dunweg3 min readEveryone asks whether AI can design the screen. For us that was never the interesting question.
Choosing was always the work. AI only changed the speed. Exploring directions for a flow used to take days. Now we do more of them in a few hours, see which ones fall apart the moment you can click them and drop those ourselves. Only the directions worth it reach the client and the user test. We explore more and commit later.
Prototyping is where it shows
The build day in a sprint used to be a pressure cooker. You picked one idea, built the one flow where nothing goes wrong and hoped people clicked where you meant them to. More did not fit in the time.
That ceiling is gone. The prototypes we build now behave like the real thing. Buttons work, inputs take real input, errors show up where they should. And we build them about three times faster than before. The speed buys room to build two directions in full instead of betting on one, and when we load the client's own design system, the prototype looks like their product instead of a sketch of it.
It changes what a test is worth. People stop filling in the gaps of a fake screen and react to the actual idea. In a recent sprint with SPIE we looked at two ways to enter large chunks of time, a quick row-by-row table and a spreadsheet import. Before, we would have prototyped one. This time we built both, and testers pasted their own spreadsheets straight in. They were testing with their own week of work. For Safeguard, a workplace-safety product, the clickable prototype is built with AI and lives online, so the people it is for can try it and we learn from how they really move through it.
What AI does not do
None of this runs on its own. AI is fast and tireless and has no idea which solution is right. It does not sit next to someone and watch them get stuck. It does not carry the weight of a decision, or read the context that makes one answer right here and wrong somewhere else. That part stays with us: which direction to back, what to build and how to keep the product's vision intact while AI does the lifting.
Clickable, interactive designs at an early stage that are nearly indistinguishable from the final product.
