Designers and AI: who decides

Why designers are starting to see AI as a colleague

In 2026 it is hard to find anyone working in digital product design who still believes artificial intelligence will simply take over their job. Figma’s annual survey on the state of the profession found that 91% of designers say generative tools improve the final quality of their work, 89% report moving faster, and 85% consider AI essential to their future career. Designer Fund reports similar numbers, describing 2026 as a turning point: most people in the field no longer frame the conversation as “AI versus designers,” but as “AI alongside designers.”

That kind of agreement is rare in a field that usually can’t settle on the right font for a button. It arrives just as a parallel trend accelerates: agentic AI, systems able to carry out tasks on their own, without a person steering every single step. This goes beyond generating an image or a rough interface on request. Agents now explore variants, test combinations, and hand back solutions already assembled and ready for review.

What changes when AI generates layouts and variants

Not long ago, the work of an interface designer often started from a blank page: wireframes, moodboards, rough sketches before anything presentable took shape. Today that phase can shrink down to a single prompt. A generative engine can return ten different layout options for one screen in minutes, suggest color variants, and surface visual hierarchies drawn from thousands of existing interfaces.

Designer Fund captures this shift with a useful image: the designer becomes an “Agent Captain,” someone who no longer produces every single artifact but oversees fleets of agents handling production. The center of gravity moves from manual execution to direction, not unlike what happened decades ago when layout software freed graphic designers from ruling grids by hand, while leaving intact the need to know which grid to pick in the first place.

UX Collective tracks the same shift. Arin Bhowmick argues that automation hits production and prototyping work hardest, the very tasks that traditionally served as an entry point for junior talent. The future of the craft, he writes, sits in orchestrating intelligent systems, human judgment and machine behavior, more than in producing static screens on their own.

Users are no longer only people

There is a second, less discussed front, raised by the Nielsen Norman Group in AI Agents as Users: for the first time, the word “user” no longer automatically means a human being. AI agents now browse sites, fill out forms and complete transactions on behalf of the people who dispatched them. Designing for them means favoring clear labels on clickable elements, predictable interaction patterns and solid semantic structure over purely visual shortcuts, which happen to be exactly the principles of digital accessibility, now also a matter of plain business sense.

There’s a neat paradox here: the more interfaces turn generative and adaptive (a trend we covered in our piece on generative UI), the more they need a solid, coherent foundation underneath. And as more agents interact with a given product, a theme we explored in accessibility by default becomes central again: not an option bolted on later, but a starting condition.

What stays out of AI’s reach

This is, to me, the real heart of the matter, more a personal conviction than an established fact. A generative system can hand back recurring patterns, statistical probabilities and combinations that have worked elsewhere in an instant. It has been trained on millions of interfaces and knows what “usually” produces good results. What it lacks, I’d argue, is intuition: that quiet form of knowledge which, as James Harrison writes on UX Collective citing psychologist Daniel Kahneman, means “thinking that you know without knowing why you do.” It’s the judgment that lets a designer sense, before being able to argue it, that a technically sound solution simply won’t work for that product, that context, that specific audience.

In my view, the creative spark is missing too, that unpredictable departure from what already exists. A language model, by design, remixes what it has already seen: it offers the statistical average of the best-known solutions, not an idea nobody has had yet. Andrea Grigsby puts it well on UX Collective, describing taste as a muscle built through experience: mistakes, trial and error, and repeated exposure to other people’s work, not something you can ask a chatbot to hand you. “Taste is a muscle,” she writes. “You can’t ask ChatGPT to give it to you, you gotta put in the reps.”

Finally, and this is where I feel strongest, there is the power to decide, a real responsibility rather than a technical footnote. Among ten variants generated automatically, someone has to pick the right one, defend it in front of a team, a client or a real person who will use the product, and answer for it if it turns out wrong. AI doesn’t sign off on that choice. It suggests, at most it refines, but the final verdict remains a human act, with all the weight that carries.

From producing to deciding: the new UX job

The practical outcome of all this is a redefinition of the role, not a shrinking of it. Time once spent producing screens gets freed up for work that demands a deeper grasp of context: understanding why a user actually abandons a flow, deciding which problem is worth solving in the first place, holding together the coherence of an entire product system. As the Nielsen Norman Group notes in State of UX 2026, pretty much anyone will soon be able to put together a decent-looking interface. What will keep making the difference is the ability to reason deeply about complex problems and their real business impact, rather than the output itself.

It’s no surprise, according to Designer Fund, that some companies which had cut creative roles to bet on automation are now walking that back: the expected savings often failed to show up, because nobody was left to make sense of what the systems were producing. We covered a similar pattern ourselves, looking at how, in certain settings, a small, tight-knit team holds up better than one gutted to make room for AI. Nor is it surprising that, as we wrote about the growing demand for humanities skills in an AI-driven world, companies increasingly need people who can read cultural context and ethical implications, not just operate a tool.

How to work today with an AI that proposes alternatives

In daily practice, it helps to treat generative output for what it is: statistical raw material, useful as a starting point and never as a finished product. It’s worth asking for multiple variations rather than settling for the first one, checking that suggested patterns actually make sense for that specific audience rather than just for the average case the model was trained on, and setting aside real time for talking with colleagues, doing qualitative research and observing the people who will actually use the product: the sources of intuition no generative system can replicate on its own.

The relationship between designers and artificial intelligence, as I see it, looks less like a contest and more like a partnership between two different roles: a tool that can explore a vast space of possibilities quickly, and a person who can decide which of those possibilities deserves to exist at all.