AI prototypes are fast; real user tests decide
Generative AI shortens the path to a prototype, but user testing decides what moves on to the product.
Generative AI shortens the path to a prototype. User testing decides what moves on to the product.
On SaaS platforms, with an international team, I used generative AI to validate ideas quickly and prototype interfaces.
In the same work, I planned and ran qualitative research and usability tests, both in person and remote.
The two went hand in hand. One does not replace the other, and that division of roles is what I write about here.
The temptation today is to skip testing because the prototype was ready early. A finished prototype is not a validated idea.
I work on products end to end, from discovery to delivery, in B2B and B2C, and I run qualitative and quantitative research along the way.
The decision here belongs to neither the designer nor the AI. It belongs to the person who will use the product, at the screen, trying to do a real task.
This piece is for anyone in a hurry to launch. A quick prototype helps with that hurry; a skipped test sends the bill later.
What generative AI speeds up
The work was a redesign of SaaS platforms, focused on usability, scalability, and standardized flows.
In a redesign, the question comes early: which idea deserves to become a screen? Drawing every option by hand takes time the project doesn't always have.
With generative AI, I validated ideas and prototyped interfaces quickly.
The gain is seeing the idea standing before investing in it. A weak option falls early, before it eats the team's time.
A quick prototype also helps in conversations with stakeholders. It's easier to discuss something you can see than a description.
But a quick prototype only shows that the idea can be drawn. It doesn't tell you whether anyone can use it.
To the AI, a button in the right place and a button in the wrong place cost the same. To the user, they don't.
That's why I treat a generated prototype as a draft for testing, not as a deliverable.
What only testing answers
On the same project, I ran research with stakeholders and users to map pain points and opportunities.
Hearing both sides adds up. Stakeholders know where the business hurts; users show where the screen hurts.
In-person testing shows the face, the pause, and the doubt. Remote testing shows people in their own setting, with their own distractions.
In both, the question is the same: can the person complete the task without help? AI can't answer that for them.
A polished prototype can hide a confusing flow. Testing exposes the flow, because people don't read the screen the way its designer does.
Before that work, I also did field research with users to validate solutions in contexts with low digital adoption.
There, I learned that users don't always have the digital background the team imagines. No generated prototype shows that.
Only by watching someone use it can you see where they get stuck, and that's where design has work to do.
From prototype to product
On the SaaS work, I also created flows, wireframes, and high-fidelity interfaces in line with the design system.
This is the stage worth investing time in, because what reaches it has already been seen working with real people.
What fails the test is not a loss. It's an idea that was cheap to drop, thanks to the quick prototype.
After launch, I analyzed metrics and feedback for continuous improvement.
Metrics show what changed; feedback and testing explain why.
Learning from the people who use the product doesn't end at delivery. AI comes back in: new hypotheses become quick prototypes, and testing decides again.
The split I argue for is simple. AI generates options, testing chooses, and the design system gives shape to what was chosen.
If you're building a product, use generative AI to reach a prototype fast. But set aside time and budget to test with the people who will use it.