Something changed about design in 2023 — and most teams haven't processed it yet.
Design used to be slow. Wireframes took days. Iteration cycles stretched across weeks. The bottleneck was production. You hired designers to move faster.
Then AI arrived, and the bottleneck moved.
You can now generate ten UI variations in the time it used to take to sketch one. Prompting a layout is faster than opening Figma. The tools that once required craft now require only a description.
Speed is solved. What isn't?
The thing that's still hard is judgment.
Knowing which of those ten variations is actually right — for your users, for your product's specific stage, for the mental model your customers already carry — that's not something you can prompt your way to.
AI has no history with your product. It doesn't know that your power users hate the sidebar you're about to redesign. It doesn't know that the checkout flow you're simplifying is the one thing enterprise buyers require to stay complicated. It cannot tell the difference between a UI that looks clean and a UI that works for the specific, irrational, habitual humans who use your product every day.
It generates. It doesn't understand.
Here's what we've watched happen in the last eighteen months.
Design output has exploded. More screens, more prototypes, more variations delivered faster than ever before. Founders see the velocity and feel progress.
Then they ship. And the problems remain. Onboarding still leaks. The core workflow still confuses new users. The dashboard still tells power users nothing useful. The support tickets are the same ones from last year, wearing slightly different clothes.
Faster production did not mean better outcomes. It just meant more of the wrong things, more quickly.
There is a word for this: aesthetic competence without product understanding.
AI is extraordinarily good at producing interfaces that look like good design. The proportions are right. The spacing is defensible. The components are familiar. It has absorbed every design system, every Dribbble shot, every UI pattern ever published.
What it has not absorbed is your product. Your users. The thing your specific customers are trying to do, and the specific ways they fail to do it. That knowledge doesn't come from a prompt. It comes from being inside the problem.
So where does AI actually help?
We use it. We're not making a romantic argument for doing things slowly. AI accelerates certain kinds of thinking remarkably well: generating the 6th option when you've convinced yourself there are only 5, producing rough structure quickly so we can critique it rather than create it, writing first drafts of microcopy so we can edit rather than invent, and stress-testing layouts across device breakpoints before we've committed to anything.
What these have in common: they compress the divergence phase. They make the space of options bigger, faster.
The convergence phase — deciding which option is right, and being able to defend that decision with specific reasoning about this product and these users — that's ours. That cannot be delegated.
The design work that actually moves products forward has always required conviction.
Not aesthetic preference. Not trend awareness. Conviction. A specific point of view about what your users need that you can back up with evidence and defend under pressure. That conviction requires knowing things — talking to your users, understanding how your product fails them, sitting with the ugly parts of the experience long enough to understand why they're ugly.
No model has done that research. No prompt contains that context.
We use AI where it accelerates thinking.
And we bring what it can't: specific conviction, for your specific product, grounded in the actual work of understanding what's breaking and why. That's the job now. It always was. It's just more obvious when AI makes everything else look easy.