The Mirror Problem: What Happens When AI Becomes Our Design Partner
Last week, I spent four hours trying to decide if I should use Figma's new AI auto-layout feature for a client project. Four hours. Not because the feature didn't work—it did, disturbingly well—but because I couldn't shake the feeling that outsourcing that decision meant something I hadn't quite calculated yet.
There's this old design principle: form follows function. But lately I'm watching form follow algorithm, and I'm genuinely uncertain whether that's evolution or drift. When Midjourney or DALL-E generates five variations of a hero image in 30 seconds, we're not just speeding up a process. We're changing which problems we solve first, which aesthetic directions feel "safe," which ideas never get sketched because the AI's output felt good enough. The tool shapes the hand that holds it, and the hand shapes what gets made.
Here's what I keep noticing:
- Designer intuition is getting faster, sharper, more immediately validated by AI suggestions
- The speed is intoxicating and I'm not entirely convinced that's good for the thinking part of my job
- Client expectations shift
- The ability to say "no, that direction is wrong" requires slowing down to explain why, which feels impossible when the AI has already shown them seven alternatives that "technically work"
I worked with a startup in Medellín in 2019—before this AI wave really hit—where we spent six weeks on information architecture. Just IA. We sketched on whiteboards, printed wireframes, watched users get frustrated, revised. It was slow and expensive and kind of unbearable. But something happened in that process: we stopped designing what we thought was elegant and started designing what actually solved the problem people had, not the problem we imagined them having.
Now? I can generate 50 interaction flows in an afternoon. I'm not sure this is the right move, but I do it anyway. The efficiency is real. The clarity feels real too, until you talk to actual users and realize you skipped the messy insight that only comes from friction.
The Role Shift Isn't Metaphorical
People talk about AI as a tool, like it's a screwdriver that happens to be smart. That's comfortable language. But tools don't argue with you. Tools don't suggest directions you hadn't considered. Tools don't train your eye to prefer certain outputs over others because certain outputs are statistically more similar to their training data, which means they're less surprising, which means they feel somehow "right" even when they're just average.
I notice this in myself. When I open Cursor or Copilot to write code, the suggestions are usually correct. Not brilliant. Correct. And my threshold for "good enough" drops every time. Why write a custom hook when the AI's version passes the tests? Why research an obscure design pattern when the generated solution is clean? I'm not sure this is laziness or just... adaptation. Maybe both.
The real question isn't whether designers will be replaced by AI. That's a boring question with a boring answer. The actual question is whether we'll be replaced by *ourselves*—by versions of ourselves who've optimized so much for speed and consensus that we forgot what it felt like to defend an idea that looked ugly at first but solved something real.
What I Still Don't Know
There's a difference between a tool that amplifies your thinking and a tool that replaces your thinking with faster versions of everyone else's thinking. I'm not confident I can always tell which one I'm using. Sometimes it feels like augmentation. Sometimes it feels like outsourcing the part of design that actually matters—the part where you sit with discomfort long enough to understand it.
Last month I rejected an AI-generated design direction that was objectively better than what I'd sketched. Cleaner hierarchy, better rhythm, smarter spacing. But it looked like five other products I'd seen, and I couldn't articulate why that mattered to my client in a way that didn't sound like designer gatekeeping. So I rebuilt it myself. It took twelve hours instead of two minutes. The client asked why I was doing extra work.
They had a point. I'm still not sure they were wrong.
