Founders ask me this every month. Is a design system worth the investment, or is it the thing you skip until Series B.
I’ve been leading design at Credit Karma for the last 7.5 years.
Here’s why I vouch for it now more than I did five years ago. If you actually want humans working alongside agents, the system has to be specced out, tokenized, with the instructions in it. Do that and the agents build with you. Skip it and whatever the agent makes is kind of garbage. It makes up new stuff. Every time. New component, slightly different, forever.
Then a year later you want to retheme the app and you have to go and freaking update every single component. You’re also burning tokens regenerating the same button over and over.
So it’s an eval layer. That’s the part I don’t think product leaders have caught up to. An agent can fail a story. It can’t fail a Figma frame.
I’ve watched an agent ignore a specced Storybook offer card and invent a new one. Then it did it again. That’s wasted tokens, but the worse part is what happens six months later when the radius changes and those cards don’t pick up the new theme.
This is also what a design system does for the humans. It makes you focus on putting the blocks together and maintaining the IA instead of everyone constantly redesigning the blocks again. If everyone, including the agent, is redrawing the card, it’s not a system. It’s a hobby.
Now, what you actually maintain. This is where I think everyone’s confused, including me some weeks.
A design refresh is the look. Color, type, radius, density. If the contract is real, that’s a theme. You change tokens, the stories still pass. Nobody invents a new button.
A design system refresh is the contract itself. New primitives, new APIs, new stories, a new thing an agent is allowed to emit. That should be rare, and it’s the expensive one.
But most teams I talk to aren’t doing either. They’re in between, permanently. Old Figma library. Half-ported code kit. A visual refresh that never made it into components. Engineers building one-offs because the system isn’t in their environment, so designers quietly stop reaching for it. Then a rebrand lands and everyone says the design system failed. It didn’t fail. Nobody ever said which of the three jobs they were doing.
You end up there by funding the brochure and starving the contract.
Two questions I’d ask before spending anything. First: why aren’t designers designing with it? Sometimes designers make the system and engineers still build their own custom components, because the component isn’t available in their environment, or it’s not built well and needs too much customization. Second: when someone does build a custom component and it performs, does it get contributed back so everyone can use it? If it never comes back, you forked. You didn’t maintain a system.
Figma is still where a new idea earns the right to exist. It just isn’t the thing you maintain as the source of truth.
The part worth paying for is small. Tokens, instructions, stories for the primitives agents will reuse, and a path for custom work to come home. I’d put a dedicated engineer on it even at 50% of their time. It’s a tax you pay for getting the advantages of building with agents. It’s not a forever project to restyle the company.
And founders, the answer isn’t “spend a year on a button kit,” but it’s also not “we’ll refresh later.” Later is how you get stuck in between.
Meta open-sourced Astryx in June: eight years inside the company, 13,000+ apps, live Storybook, CLI, MCP, and React 19. Still Beta. Theme it when you want a new look. Swizzle the parts that are genuinely yours. Send the custom stuff back through a story and a review. That’s how a refresh stays a refresh.
If you already have a real system, Astryx isn’t a rip-and-replace. You still need your own contract in code with Figma pointed at it. Cutting that to pay for the agents is how you take a tool out of their kit.
The file isn’t the job. The job is knowing whether you’re refreshing the look, refreshing the contract, or just living in the gap.
How this got written
Founders keep asking me this, so I finally wrote down the answer I keep giving. It’s from my own notes, in my own words, including the parts that aren’t polished. I use AI heavily in my product work and I’ll say so plainly. I don’t outsource the opinion, because the opinion is the only part that’s mine.
About the author
Ashwarya Subhluxmi has 12+ years of design experience in fintech, leading teams building AI products and marketplaces used by millions of people. She writes about design leadership, AI-native product design, and what changes for design orgs when agents join the team. ADPList Top 50 UX Coach, 2026. She advises early-stage founders and is based in the San Francisco Bay Area. uxbyash.com · LinkedIn: https://www.linkedin.com/in/ashwaryasubhluxmi