Build through shipping, not before it
The best systems are extracted from real product work, stabilized, and returned to the system. Every component I've built existed because a product needed it. That keeps the system grounded and prevents overengineering that looks impressive in a library and breaks down the moment engineering touches it.
Parity is where trust lives
When Figma and the coded component library drift apart, teams stop believing in the system. I treat the visual and coded sides as a single surface. Increasingly that means building systems that work with MCP servers and AI-assisted workflows out of the box, so engineers can pull from the system programmatically rather than translating from a spec.
Make the right thing easier than the wrong thing
Good systems don't spread by mandate. They spread because they remove pain. Strong defaults, shared language, clear documentation. When the system makes the right thing easier than the wrong thing, adoption follows.
Working with AI without losing the craft
I use AI heavily across my workflow, from structuring token architecture to building systems designed to integrate with MCP servers and other programmatic tooling. It has made me faster at the parts of the work that benefit from speed.
What it hasn't changed is the level of craft I hold at the granular level. AI is good at generating options. It's not good at knowing which option is right, or why a particular decision will cause problems six months from now. That judgment is the job. The tools just mean I spend more time on it.