Something strange has started happening in design reviews. We're spending more time talking about what the system should say and less time talking about what it should look like. The interface hasn't gone away — there are still screens, still layouts, still components to spec. But the meaningful design decisions are happening somewhere else, in the logic of the interaction rather than its visual representation.
This is disorienting if your model of design is "making things look right and easy to use." It's clarifying if your model of design is "understanding what people are trying to do and removing friction from their path to doing it." The distinction matters, because one of those definitions survives the shift to AI-native products and the other one doesn't.
The moment we're in is not a threat to design as a discipline. It's a purification of it — stripping away the parts of the job that were never the point.
What's being stripped away
The parts being stripped away: visual decoration, layout convention, component patterns, the familiar grammar of the GUI. These are real skills and they matter. But they were always in service of something deeper — helping a person accomplish something — and the degree to which they became the job rather than the means to the job was always a distraction. Not a malicious one. Just a gradual drift toward what's visible and measurable over what's hard and important.
The hardest adjustment for designers entering AI product work is letting go of the prototype as the primary artifact. A Figma file is a specific, bounded, testable thing. A set of prompts and responses is none of those things — it's contingent, probabilistic, and varies based on inputs you don't fully control. Learning to make design decisions without the certainty of a precise mock-up is genuinely uncomfortable. But it's also a more honest version of design. We never had as much control as the prototype implied. The prototype was always a simulation. Now the simulation is harder to sustain.
What remains — and becomes more important
What the shift to AI surfaces: intent modeling, flow design, the mental model of the user, the edge cases where the system will fail and what should happen when it does, the decision about how much the system should do versus hand back to the human. These are interaction design problems. They were always the harder problems. Now they're the visible ones.
There's one skill in particular that becomes central in AI product design that most designers underinvest in: writing. Not microcopy — the labels and error messages and tooltips — but the deeper craft of knowing what a system should say, when, and why. In a world where the interface is increasingly a conversational layer, the words are the design. The clarity of the language is the clarity of the experience. A designer who can't write well struggles to do this work.
Similarly, the concept of the happy path becomes more complex. In a traditional interface, the happy path is a sequence of screens. In an AI product, it's a quality of response — and the failure modes multiply. The system can be factually wrong. It can be technically correct but tonally wrong. It can answer a different question than the one being asked. It can be right once and wrong the next time for no obvious reason. Designing for these failure modes is design work, but it doesn't look like design work in the traditional sense. It looks like thinking about systems.
The screen isn't dead
None of this means visual craft stops mattering. It doesn't. Even AI-native products have surfaces — dashboards, settings, history views, onboarding flows — that require exactly the kind of visual thinking that hasn't changed. The screen hasn't died. But the center of gravity has shifted. The most consequential design decisions in an AI product are upstream of the visual layer, in the choices about what the system can do, what it does by default, and what it asks the human to handle.
Design has always been, at its best, the practice of making complexity legible and workable. Screens were one medium for that. Language is another. The discipline translates. What we have to let go of is the assumption that design equals visual design. What we get in return is design work that's more central to how the product actually functions — closer to the core questions of what the thing should do and how it should behave, rather than what it should look like once those decisions have already been made.
That seems like a trade worth making.