A few months ago I asked an AI tool to design a sign-up flow. It took under a minute. It looked great. Clean spacing, friendly copy, a nice illustration on the first screen.
Then I clicked through it in my head, the way a real person would.
What happens if the email is already taken? What does the screen look like before anything loads? What if someone's name is 40 characters long?
None of that existed. The flow only worked for a user who never makes a mistake. I haven’t met that user yet.
What I mean by "design with AI"
I don’t mean typing "make me an app" and shipping whatever comes out. I mean using AI inside a normal product design process: research, defining the problem, exploring ideas, designing flows and UI, and testing.
At every stage I ask one question: if this is wrong, who pays for it? If the answer is "the user," a human makes the call. Usually that human is me.
AI took away the hiding place. Now the judgment is the job.
Flows: AI draws the happy path, I draw everything else
AI-generated flows almost always assume things go right. Real products spend a lot of their life handling things that go wrong. So the AI draft is my starting point, and then I do the part I consider the actual job:
- Empty states. The first time someone opens it and there’s nothing there yet.
- Loading states. And what happens when loading takes ten seconds.
- Error states that say what went wrong and how to fix it.
- The way back. How someone undoes a mistake.
My rule, if you want to borrow it
If being wrong would cost the user something (their money, their time, their trust), a human decides. Everything else? Let AI draft it. Then look at it like a designer.