AI can now do a lot of the work that used to take designers hours.
It can create concepts, high-fidelity screens, prototypes, sample data, and design system parts. It can help test ideas and turn a rough concept into something you can click through. That's a big part of why AI product design is moving so fast.
At UITOP, we use AI in many parts of the design process too. And I think that's great.
The problem starts when people assume that a good-looking interface means good UX.
Don’t want to disappoint you, but it doesn't.
AI can generate an interface. It still can't reliably tell you whether that interface is a good product experience.
And that difference becomes especially important when teams use AI prototyping or vibe coding to move from idea to product very quickly.
The dangerous part is that AI-generated UI often looks good
Bad UX used to be easier to spot.
A rough wireframe looked rough. A half-finished prototype looked unfinished. A developer-built interface often made its weak points easy to see.
AI-generated UI is different.
It can produce polished dashboards, attractive onboarding screens, empty states, responsive layouts, and consistent components in seconds.
The result can look like a finished product.
But visual quality and product quality are not the same thing.
You can understand it when you start using the product.
An interface can look modern but navigation doesn't match how users think. The dashboard shows the wrong things first. The onboarding asks for too much before the user sees any value. An important action is hidden in a menu.
The interface looks good. The experience doesn't.
And if nobody on the team knows enough about UX to question the result, the mistake can survive all the way to production.
The “yellow sky” problem
Here's a simple example.
Imagine AI tells you the sky is yellow. If you know enough about the world, you know something is wrong. You don't need another AI to check it. You already know how to judge the answer.
The same thing can happen with AI-generated UI.
AI creates a navigation that looks fine. It puts 12 cards on a dashboard. It adds filters to a table. It creates a five-step onboarding flow. Everything looks OK.
But maybe users don't need those 12 numbers.
Maybe the filters are based on how the company works, not how users search.
Maybe users need to see the product's value before filling out six fields.
The interface may look perfectly fine, but you need UX design knowledge to see this “yellow sky” problem.
AI can give you the wrong answer with a lot of confidence. You need design experience to spot it.
AI doesn't know your users
Most AI-generated interfaces assume an ideal user.
They have a good internet connection. They understand the product. They know the terms. They read the instructions. They know where to click.
Real users are different.
They get distracted, make mistakes, skip steps and sometimes forget how things work. They come back six months later and can't remember where a setting is.
This gets harder with complex products.
Think about a CRM where someone manages thousands of contacts. Or an ERP that an operations team uses all day. Or a logistics platform where people have to make quick decisions with missing data. Or an internal tool where users don't have time to learn a new system.
This is where CRM UX, ERP UX, and internal tools UX need a different level of attention.
You can't just generate screens that look familiar. The interface has to support real work.
That means understanding why users behave in certain ways, what they need, and how they move through the product.
A product is more than its screens
AI can make a great-looking dashboard.
But what should the user see first?
What should they do next?
What happens when there is no data?
What if the data is wrong?
What if they enter something by mistake?
What if they have 6,000 records to search?
What happens when the system goes down?
These are product questions. Making the interface prettier won't solve them.
A good UX/UI designer thinks about these situations before users run into them. They understand how the screens connect and how one choice affects the next step.
This matters even more in SaaS UX and enterprise UX, where users often deal with large amounts of information and complex business rules every day.
The more you know, the better AI works
You might think AI makes design skills less important. I actually think the opposite.
If you understand your users, UX patterns, business needs, and product goals, you can get much more out of AI.
You know what to look for, what questions to ask and you can spot when something doesn't feel right.
That's why the work should start before you generate any screens.
Look at competitors. See how similar products work. Talk to users. Collect good examples. Understand the business rules and technical limits. Think through the main user flows.
Once you know that, AI design tools become much more useful. You can use it to test different ideas and see what might work.
AI can give you options. The designer still decides which ones make sense.
Because design thinking and real design experience are still what set designers apart from AI.
Vibe coding has the same problem
More and more clients come to us with products built through vibe coding.
The product already works. The screens are there. You can click through the main flows. From the outside, it can look almost ready.
Then we take a closer look.
The navigation may feel confusing. Important actions may be hard to find. Some flows have too many steps. Different parts of the product may work in different ways.
Sometimes the problem is even deeper. The product makes sense from the developer’s point of view, but not from the user’s point of view.
A UX audit helps us find these gaps and see what needs to change.
We go through the main user flows, look at how information is organized, and see where users may get stuck. We also look at the product as a whole and ask if the main structure still makes sense.
Sometimes a few changes fix the problem. Sometimes we need to rethink an entire flow.
And this is something we see quite often with AI-generated products.
AI can help you build something that works very quickly. But “working” doesn't always mean “easy to use.”
You still need a designer to look at the result and ask: Does this actually make sense for the user?
Would I make the same decision if AI hadn't suggested it?
I think critical thinking will become one of the most important skills for designers.
Designers will spend less time drawing every screen from scratch and more time deciding what should actually be there.
If you’re working with AI, you need to be able to look at an output and question the result.
You need to spot missing states, notice when a common pattern doesn't fit, think about what happens when things go wrong, understand how business rules affect user flows.
And you need to be able to say:
This looks good, but it doesn't make sense.
That may be one of the most useful things a designer can bring to an AI workflow.
Don't start with the prompt
One common mistake is starting with AI too soon.
Someone has an idea and immediately asks AI to design it.
I'd start somewhere else.
Look at competitors. Study products people already use. Talk to users. Collect useful examples. Understand the business rules. Map the main user flows. Figure out what the product needs to help people do. Then bring all that into the AI workflow.
Now AI has something useful to work with.
Use it to test ideas, try different layouts, and create several possible solutions.
After that, the designer needs to review the results.
Which ideas make sense?
What feels confusing?
What can be removed?
Do the user flows work?
What happens in less common situations?
The process looks like this:
Research - direction - AI exploration - human review - testing - refinement
This works much better than simply generating an interface and shipping it.
AI can speed up the work. The designer still guides the process and decides what works.
What this means for SaaS teams
For founders, AI makes it tempting to skip design.
I understand why.
You can build more with a smaller team. You can get a prototype quickly.
You can test ideas without spending a lot of money.
I wouldn't give that up.
I'd just bring design skills into the process before bad decisions become expensive.
In some cases, that means having a Product Designer from the start.
In others, doing a UX audit of an AI-generated product.
There are also cases where the product already exists and needs a redesign.
We've worked with products that were already built with AI when they came to us. It looked finished, but the team still needed to rethink the navigation, user flows, information structure, and key interactions.
At that point, you're redesigning the product, not just a few screens.
And that's much harder after the product has already been built.
Conclusion: Keep work human
I don't think the future of AI design is about choosing between AI and designers. What matters is how they work together.
Let AI handle more of the repeat work. Use it to test ideas, to create different states and versions, to speed up research and testing.
Then have someone with real SaaS product design experience check the result. At UITOP, we can help you review an AI-generated product, find UX issues, and rethink key flows around real user needs.
Today creating an interface is getting cheaper.
But knowing which interface to create is still hard.
Knowing when a great-looking solution is actually wrong is even harder.
AI can generate the interface. A designer still has to judge the UX.
That's the part I'd keep human.
