Article
Using AI to beat design blindness
Human design as the inspiration and the refiner, but with AI as a design assistant
5 min read
Article
Product Design
Claude Design changed how I work. No, really.
Design blindness is a real problem
There's a phenomenon I call design blindness. If you've worked in design for any length of time, you know exactly what I mean.
It happens after you've been staring at the same layout, the same components, the same screens for long enough that your brain stops seeing them clearly. Issues that would be obvious to a fresh pair of eyes become invisible. You're not being careless, your brain has simply saturated on the problem.
In a well-staffed team, this is what a colleague is for. You turn your screen around, share a Figma file in a DM, and someone with fresh eyes spots the thing you've been walking past for two days. A spacing inconsistency that's been nagging at you without you knowing why. A hierarchy problem you've become so accustomed to that it no longer registers. That informal, organic review, a quick "does this look right to you?", is one of the most underrated parts of working in a design team.
I don't always have that. And when I don't, I used to hit the wall alone.
My initial scepticism
I want to be clear about something before I go further: I was not an early AI adopter. While designers around me were rushing to integrate AI into their workflows, I was deliberately dragging my feet.
My hesitation wasn't fear of the new. It came from something more principled. As an illustrator and a supporter of independent creatives, I was genuinely troubled by how AI tools have scraped artwork and visual design from the internet without consent, credit, or compensation. And the flood of generic, aesthetically hollow output that followed had quietly lowered the bar for what "good enough" looks like. I didn't want to contribute to that.
So for a long time, I didn't. Then I found a way to use AI that felt honest.
The “cure”
When I hit the design blindness wall now, I go to Claude Design. I share screenshots of what I'm working on, along with my design system and documentation, and ask for variations that meet the project's specific product criteria. The outputs come back, I bring them into Figma, and I use them as a springboard. I'm not shipping what Claude produces. I'm using it to break the loop, to see angles I'd stopped looking for.
A recent example: I'd been iterating on an account-creation pattern for hours. I knew something was wrong, but I couldn't locate it. I'd looked at it too many times. Claude returned a layout variation I'd explicitly ruled out early in the process. Seeing it again, in a fresh context, made me realise the reason I'd ruled it out was based on an assumption that had since changed. I went back to my original rejected direction, refined it against the design library, and shipped it. Claude didn't solve the problem. It showed me where I'd stopped looking.
I'm still in control of what goes out the door. Claude is a thinking partner, not a decision-maker. But it's worth being honest: in a better-resourced team, a human thinking partner would be preferable. Not because Claude does a poor job, but because a colleague brings shared context, creative instinct, and the kind of pushback that only comes from someone who genuinely cares about the work.
The ceiling is your Design System
There's one significant caveat to everything I've described: this only works if your design system is solid.
Claude is only as good as what you give it. If your component library is a mess, your design tokens are inconsistent, and your team doesn't have agreed-upon patterns, AI will amplify that back at you. Strong, well-documented systems yield useful outputs. Without that foundation, you get noise.
This isn’t a criticism of the tool. It’s a reminder that AI in design is a multiplier, not a foundation.
What's actually changed
The practical impact has been meaningful. The time I used to lose staring blankly at a screen, or cycling through iterations without direction, has largely been reclaimed. I'm reaching the review stage faster.
More importantly, it's given me a way to interrogate my own work. When something generates a variation, it forces you to articulate why you prefer one approach over another, and sometimes the answer turns out to be habit rather than reason.
In a larger team, that sharpening happens through conversation: design crits, product manager walkthroughs, peer reviews from senior designers who aren't afraid to say something isn't working. Those interactions do more than surface problems; they develop you as a designer over time. The feedback loop is richer and more human than anything AI can replicate. I haven't stopped wanting those conversations. I've just found a way to function in their absence.
Still hesitant. Deliberately so.
But my ethical questions haven’t gone away. I’m still uneasy about how these models are trained, who gets credit, and what’s lost along the way. Those are live questions for me, and I’m not interested in waving them away just because the tool is useful.
What’s worked for me is using Claude as a thinking tool, a way to pressure-test ideas, get unstuck, or handle the admin side. The actual creative direction and final output are still mine, grounded in the systems and resources I trust.
A word about Junior Designers
One thing I want to address directly: the quiet displacement of junior designers by AI tooling.
Junior designers are not inefficiencies to be optimised away. They are the future of the discipline. They bring energy, questioning, and perspectives that AI cannot replicate, and they grow into the senior designers and design leads your team will need in 3, 5, or 10 years. If your team has the capacity to hire them, a junior designer should sit alongside AI integration, not be replaced by it.
AI tools can absorb repetitive, administrative, and mechanical work. That should free a design team to do more thinking, more learning, more of the craft that actually develops people and moves products forward. The danger is using AI as an excuse not to invest in human talent at all, a short-sighted mistake dressed up as efficiency.
My honest summary
I still haven't resolved my ethical concerns about AI in the creative industries. The questions about how these models were trained, what they consumed to get there, and who paid the price remain live for me.
What I've found is a use that keeps me on the right side of my own values: not generating visual design from scratch, not replacing craft, but using AI as an analytical layer, a gap-filler, a pressure-test, a way back in when I've lost the thread. The creative decisions, the final execution, and the overall design language remain mine.
Used within those constraints, it's become a genuine asset. Used without them, it risks becoming exactly the kind of output I was trying to avoid.