Most teams start a design system by cataloging buttons and colors. That's the visible part, but it's not the part that actually determines whether the system survives contact with a real, growing product built by more than one person.
Tokens before components
The design decisions that actually scale are the ones made once, at the token level (color palette, spacing rhythm, type scale) and referenced everywhere else, rather than redecided per component. A button and a card should pull their padding from the same spacing scale, not from two developers' independent judgment calls about what 'looks about right.'
The real test: does it survive someone else touching it
A design system is only proven once a second engineer, who didn't build the original components, can add a new feature and have it look like it belongs, without needing to ask what the 'right' padding or color is. That's the actual measure of whether a system scales: not how polished it looks in a demo, but how consistently a team can extend it under real deadline pressure.