I’ve noticed a shift in how teams are approaching interface development. AI-generated prototypes are becoming commonplace and, used correctly, they can significantly accelerate the iterative design process I’ve always advocated for.
The promise is compelling: describe what you need and get a working prototype in minutes rather than days. But here’s the catch – speed without direction is just expensive stumbling, faster.
The Right Way to Use AI Prototyping
The principle I keep returning to: iterate during development, not between releases. AI tools excel at this if you use them strategically. A designer can now test multiple interaction patterns in the time it used to take to build one. That’s powerful, but only if you know what you’re testing for.
Good UX research relies on the 5WH framework; Who is using this? What are they trying to achieve? When, Where, Why and How? Without solid user research upfront, AI simply helps you build the wrong thing faster. I’ve seen companies generate dozens of beautiful prototypes that all miss the mark because nobody bothered to understand actual user needs first.
Evidence Still Trumps Output
The real value isn’t in the prototype itself but in what you learn from testing it. AI can rapidly produce a paper prototype, a clickable mockup, or a near-final interface in rapid succession. This means you can validate concepts earlier and more frequently. But that evidence only matters if you’re actually putting prototypes in front of users.
Too many teams treat AI-generated prototypes as the solution rather than the tool. They admire the output, tweak the colours, and ship it. That’s missing the point entirely. The prototype is there to fail fast, to reveal what doesn’t work before you’ve committed development budget to building it properly.
The Familiarity Problem
AI models are trained on existing interfaces, so they naturally produce patterns users are already familiar with. That’s brilliant for ease-of-use – as I’ve said before, borrowing proven patterns maximizes usability. But it also means truly novel solutions to unique problems require human creativity. AI won’t spontaneously invent the equivalent of ‘reply to cancel’ for your specific domain.
Use AI to rapidly explore the conventional approaches, then apply human insight where genuine innovation is needed.
What This Changes
Development timelines should compress, but not in the way most organisations think. You shouldn’t be shipping faster – you should be testing more before you ship. If AI can generate a testable prototype in an hour instead of a week, that’s six days you can spend understanding user responses and refining the approach.
The bottleneck isn’t prototype creation anymore. It’s the research to know what to build and the discipline to test it properly. Those haven’t changed, and AI doesn’t solve them. What has changed is the cost of being wrong during development. It’s now so low that there’s no excuse for not iterating extensively before release.
The teams getting this right are using AI to fail cheaper and faster during development, so they can succeed on first release. The ones getting it wrong are using AI to fail more expensively by building elaborate solutions nobody actually needs.
Key Points:
- AI speeds up prototype generation, not the need for user research
- More prototypes mean more opportunities to test and iterate
- AI naturally produces familiar patterns – good for usability, limited for innovation
- Lower cost of iteration removes excuses for poor user testing
- Speed should enable more learning, not faster shipping
Since 1996, Leo has been helping organizations provide an intentional customer experience while matching technical innovations to market needs. He uses the Akendi blog to share his thoughts about the challenges of addressing business problems from an end-user perspective and finding solutions that work for real people.