Why AI Doesn't Eliminate the Need for a UX Designer
Every few months there's a new tool that promises to make UX designers unnecessary. Type a prompt, get a polished interface. Feed it your product idea, get back a set of screens. I've tried most of these tools myself, and some of them are genuinely impressive at generating something that looks like design work in a matter of seconds.
But there's a difference between generating something that looks like design and actually doing the job a UX designer does. Most of the conversation around AI and design gets stuck on the visual layer, whether AI can produce good-looking screens or match a brand's aesthetic. That's a fair conversation to have, but it skips over the part of the job that's actually harder to automate: the human-facing work. Talking to real users. Watching how they actually behave, not just what they say. Sitting in a room with stakeholders who all want something slightly different and finding a way through it. That work doesn't go away just because a tool can spit out a wireframe.
AI can talk to people, but it can't connect with them
There's a meaningful distinction in UX research between what people say and what people do, and it turns out that distinction is exactly where AI tools run into trouble.
AI is decent at the "say" side of things. It can conduct structured interviews, ask follow-up questions based on responses, and do it in dozens of languages without getting tired or bored after the fortieth repetition of the same script. If you need surveys or interview-style feedback at scale, there's real value there.
Usability testing is a different animal. Usability testing depends on watching what someone actually does with a product, not just what they say about it afterward. People think aloud during these sessions, they hesitate, they click the wrong thing and try to cover for it, they get frustrated in ways they won't fully admit to in words. Researchers at the Nielsen Norman Group have tested a number of AI research tools that claim to analyze usability sessions, and found that none of them could actually process what happened in the session itself, only the transcript of what was said. That's a real gap, because usability testing is fundamentally about behavior, and behavior isn't always something people narrate accurately, or at all.
One review of AI agents used to run usability tests on prototypes found the agents struggling with tasks as basic as clicking the right tab or finding an upload button, regardless of how detailed the instructions were. These weren't edge cases. They were the kind of everyday friction points a human moderator would catch instantly just by watching someone's face change.
There's another problem that doesn't show up in any tool comparison chart, and it's simply that people behave differently when they know they're talking to a machine. Anyone who's spent time on the phone with an automated customer service line knows the feeling: you clip your sentences, you stop explaining yourself the way you would to a person, and at a certain point you just say whatever gets you through the interaction. Some of that is impatience. Some of it is a kind of low-grade discomfort, almost embarrassment, about being candid with something that isn't going to actually understand or care. People soften their real opinions, or skip the messy explanation entirely, in a way they wouldn't with another person sitting across from them who's nodding along and asking a genuine follow-up. A user might tell an AI interviewer that a checkout flow was "fine" because typing out the actual frustration feels like more effort than it's worth. That same user, talking to a human researcher who's clearly listening and reacting, will often go on for five minutes about exactly what annoyed them and why. The gap between those two answers is the whole ballgame in research, and it's a gap that has nothing to do with how sophisticated the AI gets.
There's also the matter of what happens in the room that never makes it into a transcript at all. A user who goes quiet at a certain screen. A stakeholder who tenses up when a feature gets questioned. A participant who says the interface is "fine" in a tone that makes it obvious it's not. None of that shows up in text. All of it shapes what a good researcher does next.
Stakeholder work is a different skill than most people expect
Nobody talks about this part enough, but a huge amount of what a UX designer actually does isn't sitting alone at a desk designing screens. It's translating between people who don't agree with each other. The engineering lead who wants to ship fast. The compliance team that needs certain steps to stay in place no matter what. The executive who has a strong opinion about a button color and needs to be steered toward the actual problem. Navigating that isn't a research task or a design task exactly. It's closer to diplomacy, and it requires reading a room, adjusting your approach mid-conversation, and sometimes telling someone something they don't want to hear in a way they can actually hear it.
AI has no read on any of that. It doesn't know that the VP who just spoke up has more political capital in this meeting than the person whose title suggests they should be the decision maker. It can't sense when a stakeholder is nodding along in the meeting but is going to quietly kill the idea afterward. These are the moments where projects actually succeed or stall, and they happen in rooms, not in prompts.
The tools need more guidance than people expect
Here's something that doesn't get said enough: using AI design tools well is its own skill, and it's not a small one. Vague prompts produce vague, often wrong, results. One breakdown of AI prototyping tools found that without specific, detailed guidance, the AI's assumptions tend to miss the mark entirely, and that outputs improve dramatically when you feed the tool something concrete to work from, like an existing wireframe or a moodboard, rather than a text description alone.
This creates an odd irony. The people who get the most out of AI design tools tend to already have design training, because they know what "good" looks like and can steer the tool toward it, catch when it's wrong, and know what to fix. Someone without that background often can't tell the difference between an output that's actually solid and one that just looks confident. And AI tools are very good at looking confident. Design writers have pointed out that this is a real trap even for experienced designers: the tool hands you something polished-looking, and it's easy to stop asking whether it's actually right and start just asking how fast you can ship it.
I've seen this play out with non-designers at client companies who try to use these tools to move fast without a design person involved. The output looks like a real interface. It's laid out, it's got the right components, it kind of makes sense at a glance. Then you actually try to use it and realize the flow doesn't match how anyone would really move through the task, or a critical edge case (what happens when someone has no data yet, or hits an error) was never considered because nobody thought to ask the tool about it, and the tool had no reason to bring it up on its own.
That's the pattern across almost everything AI can currently do in this field. It's fast at producing a first draft. It needs someone who already knows what they're looking at to tell it whether the draft is any good, push it in the right direction, and catch the things it missed. Left alone with a non-designer at the wheel, it tends to produce something that looks finished but isn't.
We've been here before, just with a different tool
None of this is really new, either. Squarespace and WordPress have offered pre-built templates for well over a decade now, and plenty of businesses have used them to get a website up quickly and cheaply. Nobody in the design industry ever claimed those tools would replace designers, because it was obvious what they were good for and what they weren't. A template gets you a page that looks reasonably professional. It doesn't get you a layout built around how your specific users actually move through your specific product, or a checkout flow designed around the particular reasons your particular customers abandon their carts.
AI-generated design is really just the next version of that same idea, dressed up to look like something new. Instead of picking from a fixed library of templates, you're now generating a template on demand, one that's been assembled from patterns the model has seen elsewhere. It's faster and more flexible than clicking through a theme gallery, but the fundamental limitation is the same. It's producing something generic and adapting it to fit, rather than starting from your users' actual behavior and building something around that. A template, whether it comes from a drag-and-drop builder or an AI prompt, is still a template. It's a reasonable starting point for a lot of businesses. It's not a substitute for the kind of design work that comes from watching real people struggle with a real problem and building something specifically to fix it.
None of this means avoiding the tools
To be clear about where I actually land on this: I use AI constantly in my own work, and I'd never tell a client to avoid it. The tools that generate quick variations or spin up a rough prototype in minutes instead of hours are genuinely useful, and pretending otherwise would be its own kind of dishonesty.
Where it helps most for me is speed and accuracy in the parts of the process that used to eat up time without adding much insight. I use AI to build out prototypes faster and get something more accurate in front of people sooner, which changes the quality of everything downstream. Stakeholders approve directions faster when they're looking at something that already feels close to real, instead of a rough sketch they have to squint at and imagine. Usability testing gets better when participants are reacting to a prototype that behaves the way the real product will, rather than a static mockup that requires constant explanation of what would happen if they clicked something. Developer handoff goes smoother when the specs and states are more fully fleshed out going in, instead of getting figured out mid-build through a string of back-and-forth questions.
None of that replaces the parts of the job this article is actually about. It just means the drafting and iteration happen faster, so more of my actual time goes toward the interviews, the testing, and the stakeholder conversations that AI still can't run on its own. That's the honest version of "AI and UX design," at least in my experience. It's not a replacement for the person. It's a way for the person to spend less time on the mechanical parts and more time on the parts that were always the actual job.
What this actually means
None of this means AI is useless in design work. It's genuinely useful for early exploration, for generating quick variations, for handling some of the more repetitive parts of research at scale. But the parts of the job that involve real judgment, sitting with actual users, reading a room full of stakeholders with competing agendas, knowing when an AI-generated draft is actually solving the problem versus just looking like it is, still need a person who's done this enough times to know the difference.
If your team is trying to figure out where AI genuinely helps and where you still need someone steering the process, that's a conversation worth having before you build a product roadmap around a tool that can't tell the difference between a design that works and one that just looks like it does.
Ursa Design is a fractional UX design practice for product companies and IT consultancies. If your team has design coverage needs, let's talk about what fractional design could look like for your team.
Contact
Reach out today and you’ll get a clear plan, honest advice, and a design expert that cares about the outcome as much as you do. Whether you prefer a quick call or a simple email, getting started is easy.