AI can build your UI now — so what's the first thing you'd still want a human for?
The baseline has moved fast this year. AI-generated UI is good enough now that for a lot of products the core screens come out clean on the first pass , dashboard, settings, checkout, all serviceable. That used to be weeks of work for a solo maker. It isn't anymore, and I don't think that's a bad thing.
I'm a UX designer, mostly on complex, high-trust products, and here's where I think it currently stops: AI generates screens, but product experience lives in the sequence between them. The happy path is basically solved. What doesn't come out of a prompt is what happens when someone is confused, hesitant, or something broke — empty states, the permission ask that feels invasive, the error that doesn't say what to do next, the ability to undo something. None of those are visual problems, so a tool optimizing for a good-looking screen has no reason to raise them.
That's a designer's view of it though, and I'm aware it's a convenient one for me to hold. So I'd rather hear whether it matches what people building solo actually run into:
• The last time a user got confused in your product, where was it? A specific screen, or something between screens?
• If you've used AI for design, what did you end up redoing by hand? And was that because it looked wrong, or because it behaved wrong?
• Has anyone brought someone in for one specific flow rather than a whole redesign — onboarding, checkout, permissions? Did that work, or did it need the wider context to be useful?
Genuinely curious where I'm wrong on this. The "AI is enough" answer might just be correct for most products and I'd like to know that.
Replies
This matches my experience pretty closely.
I’m building SteadFolio as a solo founder, and AI has made the actual UI production dramatically faster. Getting a clean dashboard or a functional screen is no longer the bottleneck.
The harder part has been everything around it.
For example, onboarding. The individual screens can look perfectly fine, but the sequence can still feel too long, confusing, or ask for something before the user understands why it matters.
The same applies to empty states and errors. AI can generate a technically correct error message, but that doesn't necessarily mean the user knows what to do next.
I’ve also found that the biggest improvements often come from watching someone use the product rather than looking at the screens myself.
So I agree with your point: AI has made screen design much cheaper. It hasn't made product thinking cheap.
And honestly, that's probably where a lot of solo founders underestimate the work.
@andreas_broutas Ha, "AI made screen design cheaper, not product thinking", better than how I said it. Taking that one.
Empty states are the one I'd underline from your list. They're the first thing every single user sees, and they almost always get designed last, usually from a mockup that's full of fake rows. So the version that gets built is the one with twelve items in it, and the version people actually land on is blank.
Which means the empty state ends up doing the job of onboarding whether you planned it or not. It has to say what goes here, why it's worth the effort, and give one obvious thing to do next. None of that is layout, so nothing prompts for it.
BetterClaw
Homepage and banners
@better_shaya Interesting — that's the opposite of where I'd have guessed. Makes sense though: a dashboard can look like every other dashboard and nobody minds, but the homepage is the one place where looking generic actually costs you something.
Is it the visual side you'd want help with, or the working-out-what-to-say part?