1. Shipping the first output
AI gives you something that looks finished. That's the trap. It's not finished, it's a draft that happens to have drop shadows. Treat it like a sketch and you'll be fine. Treat it like a deliverable and you'll be redesigning in two weeks.
2. Forgetting that accessibility doesn't come included
The models are trained on the internet. The internet has terrible contrast ratios and no focus states. So when you generate a UI and think "looks good," run it through a WCAG checker before you feel good about it. Every time. Without exception.
3. Three AI sessions, three visual languages
You generated some screens on Monday, more on Thursday, handed off to someone who generated the rest on Friday. Now your app looks like three different products. AI has no memory of your brand. You have to bring that context every single time, or it'll invent something new.
4. I asked AI what users want
AI has never used your product. It's never been confused by your onboarding, never rage-quit your checkout flow, never squinted at your mobile screen in direct sunlight. It can help you analyze research. It cannot replace doing it.
5. Prompting for vibes instead of problems
"Make it modern and clean" is not a brief. AI will produce something modern and clean that solves nothing, because you didn't tell it what problem you're solving. The more specific your context, the more useful the output.
6. Copy-pasting microcopy without reading it out loud
AI microcopy is grammatically correct and completely lifeless. Read every label, tooltip, and error message out loud before it ships. If it sounds like it was written by someone who has never had a conversation, rewrite it. Takes five minutes.
7. Accepting AI-generated navigation as information architecture
AI will give you a nav structure that looks reasonable. That's different from a nav structure that matches how your users actually think. IA comes from user research, card sorting, tree testing, actual conversations. AI can help you brainstorm options. It can't tell you which one is right.
8. Designing for nobody in particular
AI optimizes for the average case. The average case is a fiction. Your users are older, younger, slower on mobile, using a screen reader, in a hurry, distracted. Push your prompts with real constraints and specific contexts, or you'll get interfaces that technically work for everyone and genuinely work for no one.
9. Nobody reviewed the ethical stuff
AI doesn't know it's designing a dark pattern. It'll hide the cancel button, pre-check the marketing opt-in, and bury the pricing, not maliciously, just because it's seen those patterns a thousand times and they look totally normal. Someone needs to look at every flow that involves money, consent, or defaults with actual judgment.
10. Skipping the "something feels off" conversation
Fast feels like progress. So when something is slightly wrong, the hierarchy is confusing, the spacing is weird, the flow doesn't quite make sense, teams push through because stopping feels like losing momentum. It's not. Catching it in review is always cheaper than catching it in production.
AI is a genuinely useful tool. It's just not a replacement for knowing what you're doing, and knowing when to push back on what it gives you.
