We killed the design system. The product got better.
We spent three weeks building a design system for a product that had four screens. Then we deleted it. Here's what we learned about when a system earns its keep — and when it's just procrastination with a Figma licence.
The system we built
Early on a web build, we did the responsible-looking thing: we stopped and built a design system. Tokens, a component library, a documentation site, the works. Buttons with six variants. A spacing scale nobody would ever exceed. It felt like engineering maturity.
It took the better part of three weeks. The product, at that point, had four screens.
What it actually cost us
Every new screen now had to route through the system. Need a slightly different card? File it as a system change, update the docs, then use it. The system became a tollbooth between us and the work. Worse, half the components were speculative — built for screens that didn't exist yet and, it turned out, never would.
We were maintaining an abstraction of a product instead of the product.
What we did instead
We deleted the docs site and the component library. The source of truth became the actual components in the actual repo — the ones rendering real screens. When two screens needed the same button, we extracted a component. Not before. The abstraction followed the duplication instead of preceding it.
Tokens stayed. A dozen CSS custom properties for colour, spacing, and type is not a design system — it's just not hard-coding hex values twice. That part always pays off.
What got better
Velocity, obviously. But the surprising win was coherence. When the components live in the screens, you see the inconsistencies because you're looking at the real thing, not a swatch board. We caught more visual drift reviewing actual pages than we ever did in a component catalogue.
When you do need one
Three or more teams touching the same surface. A real multi-brand requirement. A component used in genuinely dozens of places. At that scale a system stops being overhead and starts being leverage. For a four-screen product with a team of three, it was procrastination with a Figma licence. We ship the product now and extract the system when the product earns it.