The training ground problem
On Figma, AI, and who builds the next generation of design judgment
Someone dropped a link to the new Claude Design announcement in a design community I’m in, and within minutes, the thread was alive. The first reply was confident that Figma would be gone in under a few years. I wasn’t ready to go that far, but I couldn’t dismiss it either.
The fear came first, the speed, the quality, what it means that a good prompt now produces something real without us. Then the dependency anxiety. If we build our whole practice around Claude, they charge whatever they want. That part doesn’t get talked about honestly enough.
Then it got more interesting.
Why Figma specifically is in trouble
It’s not just that AI got good. Figma’s format was never built to be worked with programmatically. It’s largely locked down and painful to automate. Meanwhile, agents work in code. The source of truth is migrating back to where it started, and Figma is caught holding a system nobody would design from scratch today.
Figma is already a bottleneck for agents. Claude Design doesn’t pretend otherwise. It’s HTML and JS all the way down, eventually talking directly to Claude Code.
Config this year will be interesting. Figma has announced a lot over the last few years and hasn’t always delivered. Now the ground has shifted faster than anyone expected.
The part nobody answered
The conversation shifted from fear to something more useful. The thinking layer survives, someone said, tools change, designers adapt, the profession moves forward. I’ve heard this before, and I mostly believe it.
But execution is where judgment actually forms. Not through frameworks or process through making things badly, repeatedly, until something clicks. Bad spacing decisions. Layouts that felt right and looked wrong. Iterations that taught you things no methodology could. If AI handles that layer before the next generation gets to touch it, what exactly is the thinking layer built on?
Design thinking is a process you can learn. Judgment is quieter than that. Knowing something is off before you can explain why. One doesn’t automatically produce the other.
And honestly, I don’t want to spend my career becoming a prompt generator for something doing the actual thinking.
The counter-argument was loops, small self-directed cycles, generate contrast, let signals emerge. For juniors specifically, you’re not looping against judgment early on. You’re looping against comparison. Does this feel clearer than the last one? Reps compound over time.
It’s a thoughtful answer. But I’ve watched designers do the reps and still not develop taste. Reps are necessary. They’re not sufficient. The missing ingredient is harder to name. It’s something closer to genuine attention and caring about the outcome than technique or methodology.
Nobody’s cleanly answered that part yet.