Design systems need evals
You've taught agents your design system. You still don't know whether they're following it.
Writing about design, systems, and the organisations that build them.
You've taught agents your design system. You still don't know whether they're following it.
Authored change has outpaced the review model, and that breaks more than it looks.
And whether the next generation of practitioners will get the same thing.
Reopening every decision you left implicit, and settling it without you
The practices exist. The discipline hasn't caught up
AGENTS.md, SKILL.md, DESIGN.md, and where each one fits
How the promises your system makes became the thing agents read first.
The rule design systems use to promote patterns stops working the moment AI is generating the instances.
Design system teams are governing AI-generated output – a mandate nobody assigned them. Here's what that work actually involves, and how to make it visible to your organisation.
Reading a system you didn't build, and why the first 90 days look nothing like building
The most useful feedback your system has ever produced is on a thirty-day timer
Why every layer name, variant, and token reference is an architectural decision
How your design system vocabulary becomes AI context
The gap between what your system says and what it does used to be a minor annoyance. Now it's a structural fault.
Four open-source systems. One component. What the implementation choices actually say.