BMW Design System
Same rules. One source.
I led a six-person design team on the BMW app and shaped its design system, working across vehicle capabilities, market restrictions, and light and dark modes.
A look back at what we decided in 2020, and a hypothesis for how I’d run the system with AI today.
- 2020
- Check each model in Excel
- 2026 · Proposed
- AI flags edge cases. Team decides.
Every model.
Every market.
We tracked which controls each model and market allowed in Excel. I filtered it row by row: if the law didn’t allow a control, I left the button out.
Cars respond slowly, and a car without signal can’t respond at all. We tested how long the app could wait. When a request failed, the car couldn’t say why, so the app could only tell the driver it hadn’t worked.
AI compares.
The team decides.
This part is a hypothesis, not something we shipped. I would store the model and market rules as data, linked to the Figma component and its code. AI would compare every case against those rules and collect real examples that no rule covers yet.
The team would decide what each case should do. Each decision would become a rule the next review starts from.
Edge-case report
In the system
No rule yet
- One component, linkedFigma ↔ code via Code Connect
- Rules as dataModel × market → button shown or hidden
- Edge-case scanCollect real cases no rule covers yet
- Team decidesEach new rule goes through a pull request
