Two years ago, every designer I knew was panicking about AI. Today? We’re all using it, but not for what you’d think.
What We Thought Would Happen
Remember Uizard claiming it could “do 80% of a designer’s job”? Or those LinkedIn posts about designers becoming obsolete by 2025?
I spent three months in 2023 learning prompt engineering. My colleague started a YouTube channel about “AI-proof design skills.” We were all preparing for unemployment.
What Actually Happened
The tools sucked.
I mean, they really sucked. By April 2024, our team had tested every major AI design tool. The verdict? They couldn’t even generate a decent contact form. Buttons would randomly change colors between screens. Error messages read like robot poetry. One tool kept putting navigation menus at the bottom of desktop sites.
At Germain UX, we tracked real users trying to complete tasks on AI-designed interfaces. They’d get confused at the weirdest moments—like a checkout button being green instead of the expected blue, or form fields appearing in illogical orders. Users couldn’t articulate why, but something felt “wrong.”
The Stuff That Actually Works
Here’s the plot twist: the useful AI tools aren’t trying to be designers.
Take Figma’s new features:
- Rename Layers: Turns your mess of “Rectangle 47” and “Frame 23” into actual organized layers
- Find More Like: Instead of digging through 50 files for that one button style, find it instantly
- Rewrite This: “Lorem ipsum” becomes realistic placeholder copy in seconds
Notice a pattern? These tools do the boring stuff we hate.
A Real Example: The 30-Minute Landing Page
The AI-Tamago project showed what’s actually possible: a landing page in 30 minutes using Vercel v0. Sounds amazing, right?
But here’s what they don’t mention in the hype: it only works that fast when you’re both the designer AND developer. The moment you add another person, you’re back to Slack messages, Figma comments, and “can we make the button more… clickable?”
I tried the same approach with my team. That 30-minute miracle turned into a 3-day project. Why? Because AI can’t participate in a design critique. It can’t explain its choices. It can’t push back when the PM wants to add “just one more field” to the form.
The 500-Character Problem
Want to know why AI can’t design? Try explaining your entire project in a prompt:
“E-commerce site for sustainable furniture targeting millennials must work with our design system follow WCAG guidelines integrate with Shopify consider mobile-first but also look good on desktop need trust signals for first-time buyers…”
And you haven’t even mentioned the business goals yet.
Even Figma’s First Draft (which technically has no character limit), gives you a text box that shows three lines. You know what happens when you can only see three lines of what you’re typing? You write three lines. Then the AI builds exactly what you’d expect from three lines of context: generic garbage.
What You Can Actually Use AI For (Daily)
Here’s a basic starter AI toolkit:
Every day:
- Figma’s rename/organize features (saves 20 minutes of tedious work)
- Copy generation for prototypes (no more lorem ipsum)
- Color palette suggestions when stuck
Weekly:
- Midjourney for mood boards and placeholders
- ChatGPT for writing UX microcopy variations
- GitHub Copilot for repetitive CSS
Never:
- Complete design generation
- User flow decisions
- Final production designs
- Anything involving real user data or sensitive contexts
What’s Coming (And What to Do About It)
The next wave is already here: Coframe adjusts images and copy based on user behavior. Vercel’s AI picks components based on context. But notice… someone still has to create those images, write that copy, build those components.
My advice?
- Learn the narrow tools now. The ones that save time on specific tasks. Skip the “design everything” tools, they’re not ready.
- Document your design system. AI works best with clear constraints. The better your system, the more AI can help.
- Focus on what AI can’t do. User research. Stakeholder management. Creative problem-solving.
- Track everything. We use Germain UX to monitor how users interact with our designs (AI-assisted or not). Data beats opinions, especially when everyone’s hyping AI.
