AI tools are changing how quickly product teams can move from an idea to a working prototype, but faster prototyping doesn’t necessarily mean better product design. As AI-generated interfaces and “vibe-coded” prototypes become more common, product leaders and designers are asking important questions: How useful are AI-generated prototypes? Where does AI-generated UI fall short? What should designers do differently when AI can handle more of the production work? And how can teams use AI without sacrificing usability, accessibility, or product strategy? We asked a few Fuzzy Math designers to share what they’re experimenting with, what they’ve learned, and where they see AI creating new opportunities – and new challenges – for product design.
What tools have you been exploring recently, what have you enjoyed playing around with? Anything surprise you?
I’ve been mostly using Claude and recently tried KlingAI to create short videos. Claude is probably the best LLM I’ve used so far; it can be used to create many different types of assets at a better quality than some other tools, which has been surprising to me. You still need to be quite strategic about prompting to get the best result but it’s been interesting seeing what Claude can do, especially using Claude Code with the Figma MCP connection. KlingAI is my first experience with video generation and although it can take a few tries to generate something that doesn’t show someone with three hands, it’s been quite surprising how good it can do with short videos.
I have an attachment to Midjourney, maybe because it was one of the first image generation tools. It has a reputation for not being very good, which I’ve found both silly and entertaining; however, I will say it’s stepping up its game. It’s quietly improving and recently introduced new customization features that make it easier to wield.
What’s something you do differently as a designer now that you have AI as a tool?
I definitely use it as search engine replacement. I’ve found it’s really useful for anytime I need to comb the internet for something, whether that’s articles about a specific design topic or research or UI inspiration. I’ve found a lot of the time Claude can pull up more accurate results than Google can. It’s also pretty good at writing copy, so I’ve used it to replace lorem ipsum a lot unless I think having lorem ipsum would be more useful.
Similar to Dylan, I’ve been using Claude with the Mobbin MCP to efficiently source design inspiration from real websites and apps. It’s saved me a lot of time compared to scrolling through multiple UI inspiration sites when I’m looking for examples of a specific pattern or user flow.
I’ve been working smarter, not harder, with AI’s help. I’ve always liked having a hand in every part of the process, but AI has deemed that unnecessary. I use it for narrowing down research and generating example copy, as well as for creating image assets that used to take significant time tracking down.
What’s your honest gut reaction when a client says “we already have an AI prototype”? Is there something you wish clients understood before they come to us with something AI already built?
It depends on the client. If a client is approaching us with an inherent suspicion of design’s value, I get nervous that they expect us to paint a broken canvas or frankenstein a platform with dissonant pieces.
If you’ve ever been to therapy, you understand that the true cause of a problem often requires a bit of digging through asking questions. The same is true about user experience design. I wish these clients understood that AI’s output is only as good as the context provided, which is why we spend weeks investigating a client and their users to uncover the truth of what they need before we start designing.
My gut reaction is honestly apprehension. Some clients create a prototype as a jumping off point just to illustrate their initial ideas and others create a prototype as a visual representation of exactly what they want the platform to be like. The latter is more difficult to approach because they have so much stake in what they’ve already created and design is meant to be an iterative process, where the first thing we think of is often not the final product. I wish clients understood that there’s a lot more to the design process than just producing the final screens and that process can be very difficult to replicate with AI. AI is a tool, like any other, to aid in our workflow but not replicate it entirely. I hope they understand that, as good designers, we’re going to want to iterate and improve upon their prototypes for the best user experience possible, instead of taking them for face value.
I get a bit nervous that they will be overly attached to it. Whether the design the AI has come up with is good or bad, a big part of design is the iterative nature of the process — even good ideas have to adapt and change to reflect new research or user needs in the discovery process.
Often times, before AI, there would be 20 iterations of a design that would get thrown away or recycled into something different and better before a client even clapped eyes on it.
Feels like a massive reason to keep designers in the process because we are trained to be iterative and typically have no problem scrapping what doesn’t work, and moving forward with what does. Something AI models and non-designers both can sometimes struggle with when dealing with product design.
Where does AI generated UI typically fall short? What’s the most common thing you end up fixing in an AI-generated prototype?
I think AI falls short on innovation and creativity. Sometimes it feels like AI is going for the “low hanging fruit” instead of producing the best version of something. The output can often be generic and it can look very flashy or visually appealing but function poorly from a UX perspective. The most common fixes I make are to the layout or IA of a page, often having to modify how the content is displayed.
The UI is typically generic and copied from similar websites/platforms. Those platforms might work really well, but they have different users, use cases, and branding. As a visual designer I end up making a lot of changes to properly apply a client’s branding, including (but not limited to) making the color palette and layout accessible for users with visual impairments and screen-readers.
A lot of stuff typically needs to be fixed, either by doing it by hand, or by giving the AI model better, more explicit prompts. A couple examples that come to mind:
1. If you give AI project context, that context typically includes user journey info. It usually goes something like, and I’m simplifying here, but ‘user does A, then user does B, then user does C, then user either does D1 or D2, then user does E1 or E2’ and so on.
AI sees this and will make screens into a ‘wizard style’ step-by-step linear user flow, and it will make those steps into navigation elements. Even if you, for example, asked it to make you a dashboard.
2. A big thing I see AI doing, and have for a long time, is the models will be very keen to attempt to use copy within the screens themselves to explain why design choices were made. So a simple example of this could be instead of an H1 on a dashboard saying ‘Hello, User!’ it will say “3 items require your attention, everything else handled automatically!”
The UI should be able to speak for itself and aspects of the UI should communicate to the user that 3 items require their attention and that the rest of it was handled, not the H1 copy.
What have you learned recently that has helped you produce better design-related outputs with AI?
Just as AI is learning how to think like a human, sometimes you have to think like a computer.
When we need to create high-fidelity prototypes in AI tools, I’ve been experimenting with “anti-slop” skills with built-in UI guardrails. They teach and guide AI coding agents to produce less generic, more professional front-end user interfaces, which reduces the number of refinement prompts we need to make.
One thing that is small but has been pretty impactful is that I am now in the habit of writing documents for Claude to explain the other context documents and screenshots within a project. This helps it orient and understand better when I prompt something like, ‘use Document X to search for concepts and UI that has already done what is described in the document.’
I’ve learned how to develop more effective prompts to produce better outputs with AI. I read that AI is best when treated like a design assistant instead of a designer itself and that’s been helpful in setting up AI tools for success. I’ve learned how to give more thorough context to orient the tool better for what I’m trying to achieve. Sometimes I provide a loose plan on how it should best execute, like explicitly directing it to conduct research before producing anything. I also provide guardrails and call out what I do not want the tool to do when it’s generating an asset. These things create a healthy amount of work upfront but really help with the end result.
How has the role of designer changed as more people are using AI tools? What new opportunities does this create?
I think AI gives us the opportunity to help people embrace design thinking even more, as people realize that the discipline and thought that comes from design thinking helps make better experiences and products.
Because AI is a tool, right? An impressive tool, but still a tool. If you’re not skilled before you start a woodworking project, it doesn’t matter how impressive your tools are, a table you make is still going to be wobbly.
So it allows us to focus on saying – ‘OK, you’ve got AI screens’ – let’s use those as a starting point and dive into how we can make a whole experience that works for your users.
In some cases our role has shrunk into overseers instead of producers, but in others we’re able to expand. We’re lucky to have a variety of clients here at Fuzzy Math; the big companies with teams of researchers, designers, and developers need an outside perspective to identify where they’re falling short rather than designing the solution, while the smaller companies with teams of two or three need a product designer, brand designer, and developer all in one (made possible with all the new tools at our disposal).