How AI is changing (the way I do) UX
Disclaimer: I have been working as a web designer, frontend developer and UX designer for the last 10 years, and 3 years at Spatie. I have experimented with many tools trying to make my life as a designer easier in communicating work to clients.
The first time I saw AI build a fully interactive prototype with real data, multiple flows and edge cases built in (what felt like) minutes, I understood both why this is exciting and why it's dangerous. For a client, it feels like watching every idea coming to life in real time.
I am not the only one who has seen the way they work change drastically over the last 6 months. This is not a post about the impact of AI on business, or criticism of how AI is coming for every knowledge worker.
I give you three statements of how I have seen AI change the way ideas and code are generated in my field of UX, and how I experience this change in both positive and negative ways.
1. Clients generating AI mockups, prototypes and specs improves the lives of UX designers.
With AI, I have seen some of our clients regularly go beyond just writing what they want in text. They brainstorm and rubber duck with AI before coming to us to present the work they did. And in many cases, I approve of this new workflow. It can fix the back-and-forth dance working with remote clients. It solves the 'blank canvas' problem and gives something that users of a product can look at and pinpoint issues with. However, it can also be dangerous.
These AI prototypes live in a bubble. The client sees a polished flow and thinks we're nearly there, but we've only modeled the happy path. They may say "we'll figure out the gaps later", but edge cases, error states, empty states, loading states, are often where real UX issues live, and they're invisible in an AI-generated prototype unless you specifically design for it.
At Spatie, our clients are generally technical enough to understand that what's generated can't be used 1:1 in a real application, but I know of agencies where that distinction is harder to communicate, and the more convincing the prototype, the harder that conversation gets.
2. The abundance of component and pattern libraries makes for safe and boring UX and UI.
This is a tricky one and depends a lot on what you're building, who you're building it for, and why. Practicing safe UX is the right choice for critical flows like checkout, where you want users to have a sense of familiarity. You don't want users to figure out your checkout when they're trying to pay for their new summer outfit.
But AI likes common patterns, meaning the more specific your problem, the more chance you'll get a confidently wrong starting point. The better instinct is to draw inspiration from unrelated systems, physical experiences, other industries, rather than prompting your way to a solution. It could entice you to explore outside of commonly thread paths.
Most of my ideas come from just paying attention. On a recent trip to Japan, I saw how Tokyo's metro uses letters for lines and numbers for stops, no more memorizing a station name, but only remember to get off at "JY16". This solves a language barrier and helps the types of people who count their stops. While these kinds of things might not be applicable to many of your own problems, having your mind full of little experiences (positive or negative) makes you a better designer. No component library has those things defined as patterns.
3. Research, qualitative user testing and gathering actual usage data will never be replaced by AI.
A big part of my job is to understand problems. Before I can understand a problem, I need to understand how a business operates and works inside their industry. That means talking with people, not only to gather resources and insight, but also the physical act of talking, showing genuine interest. Making clients feel heard and their challenges understood is going to matter more than ever.
Quantitative user testing (or testing hypotheses at scale) has been (mostly) automated before AI. But qualitative testing, sometimes just for one or two sessions, means watching users closely: noticing the sticky notes on their monitors, the personal spreadsheets they've built, the printed checklists that tell you exactly where friction lives, especially in early phases of gathering those insights.
That said, AI can be a genuine thinking partner when there's research it can fall back on. For one large project, we gathered glossaries, feature descriptions and user expectations into a shared system before involving AI at all. With those in place, AI could work from real knowledge rather than generating plausible-sounding assumptions.
Messy, human problems
None of these statements are arguments against AI. I use it daily, and it's made parts of my job faster, more collaborative than before. But the parts of UX that were always hardest to automate are now the most valuable.
AI doesn't sit with messy, human problems long enough to truly understand them. That's always been the job. And the designers who keep doing it well will be the ones who matter most.
