Design and AI Collide
Anthropic released Claude Design this month and it caused quite a stir in the tech industry. The storm front that hit software engineering last year has moved its way over design, looming with its friendly helpful posture. What made Claude Design more ominous than other AI design tools, is the insane pace in which the Anthropic team works. What we see and use today, certainly will become more powerful and disruptive over the next 6-12 months.
I played with it for a few days. The experience went the way it usually goes for me when a new AI tool drops. The initial feeling is amazement but a day passes, maybe two, and the amazement fades into something emptier. What do I actually do with this? How do I make something I really like? Reality set in when I had to go through a real problem and try and solve it. I got frustrated as soon as I had it make create something exactly as I envisioned it. I found myself telling the LLM to nudge the box 4px up and then left. Then I ran out of tokens and rolled my eyes.
This triggered something in me that most designers know well already. The shininess of the artifact is a powerful output that is easily valued by non-designers, but it's far from everything we do. With tools like Claude Design, what happens when the artifact gets automated? The insecurity none of us want to admit peeks out and that's the thing to pay attention to right now. I'm here with a reminder if you need it: there is so much more to your work than the artifact.
What AI doesn't touch in the design process is the part that anchors the artifact in the first place. The ability to organize and combine inputs (research, constraints, customer needs, business outcomes, brand, context) in such a way that something new and unique forms through trial and error. The judgment about what to amplify, what to kill, what to try next. The taste that comes from doing the work enough times to know what works for these people in this context. Then putting it in front of real people, watching how they use it, and changing what doesn't work without ego.
That's what I'd call the dark matter of design. It doesn't appear in the deliverable but it shapes what the deliverable becomes. The artifact is about to get cheaper. The judgment that makes it worth anything is not.
History Rhyming
We saw the same dynamic play out for engineers already. Coding tools went from tab completion to writing the majority of the code in a given day. The headlines went straight to "SWE is dead." Six months in, engineering dark matter has become more central to the work. How you build systems that scale? How you keep them maintainable when someone a year from now has to extend them? How you organize the inputs to AI, and how you judge the outputs to keep quality from drifting? These are some of the skills that make great engineers great. The syntax was always downstream of them.
And we're starting to see what happens when those skills weren't applied in the rush. The "year of slop" is what happens when code is shipped fast by people learning to code with AI. It has real consequences and companies are realizing that the hastily shipped code is brittle and hard to extend. Debt is accumulating in ways that will take skilled craftspeople weeks and months to untangle. We're finding that the engineering dark matter wasn't optional. The result is that demand for AI powered engineers is on the rise not a decline.
Let that be a lesson for design. The tools are about to get strong enough to produce a lot of artifacts, fast. If we let artifact production happen without the dark matter in the room, we're going to enter our own year of slop. And ours will look different from engineering's.
Design slop is a race to the average. There's already a gravity in the field toward the best. Apple, Stripe, Linear, Airbnb are teams that have spent years building taste through their dark matter. They put the customer at the centre. They build apps that get out of the way. Their visual language sets the bar. The full package. AI is trained on what good looks like, so when the "vibe design" happens without judgment, the output drifts toward those references, and not in a good way. Everyone's work converges. Same sans-serif. Same restrained palette with one accent color. Same rounded corners, soft shadows, light/dark toggle. Gorgeous. Indistinguishable. Everything could be anything.
And then someone is going to have to undo it.
Light Up the Dark
The one good thing about AI automating artifact creation is that it changes which skills are visible. The dark matter has been doing the work all along, but it's been hard to see because the artifact was sitting on top of it, taking the credit. With artifact production moving to AI, the skills underneath get to shine. The work that always made design valuable is the work that's about to get the spotlight.
Visibility won't come automatically. Dark matter only gets credit if the people paying for design can see it. If a designer's value is the judgment behind the artifact and the artifact becomes cheap, the work of making that judgment legible to founders, PMs, and execs becomes part of the work. We're going to have to advocate for the dark matter out loud.
I worry it will take going through a rough transition. The market, the businesses, the design teams may have to make the mistake and heal from it before the lesson sticks. It becomes earned scar tissue. The reminder that good design was always more than a Figma file. The Figma file was a small representation of the hidden work that made the design valuable in the first place.
How It Shakes Out
A few practical thoughts to close on. Claude Design is a great step forward. It feels to me like it's in roughly the state Claude Code was in a year ago. An interesting experiment that's worth playing with. It generates ideas faster than I can on my own, which becomes a tool for me to explore with. In its current state it isn't usable for professional work yet or replacing Figma. I can barely generate a site before my tokens run out. What it does do, is buy us time to try, test and learn. Experiment and find your path with the tool. Elevate your vision in ways you couldn't before. Let it ignite your dark matter skills.
Looking ahead a year, if I've learned anything about 2025, it's that what we have today will change and evolve. Maybe we get another December surprise and what happened to Claude Code happens to Claude Design and the world changes again, overnight.
The future of design will not belong to the people who can generate the most screens. It will belong to the people who know why a screen should exist at all.