The Shift
Two years ago, design systems were about efficiency - building a component library so designers didn't reinvent buttons. Today, with AI generating UI in seconds, the efficiency argument has changed. The question isn't "can we build this faster?" - AI handles that. The question is "does this feel like us?"
That's the new job of a design system: encoding brand identity and design decisions so that AI-generated output stays on-brand. It's a guardrail system, not just a component library.
What You Can Safely Automate
Component generation from tokens is now trivially automatable. Give AI your spacing scale, color tokens, and typography ramp, and it'll generate button variants, card layouts, and form elements that are structurally correct. Same for documentation - AI can write usage guidelines, prop tables, and accessibility notes faster than any human.
What Still Needs Humans
Token architecture: How your spacing, color, and typography systems relate to each other. This is the DNA of your visual language. AI can't decide whether your brand uses a 4px or 8px base grid, or whether your primary action color should be warm or cool.
Brand coherence: AI generates components that are technically correct but emotionally inconsistent. A button that passes every accessibility check can still feel wrong next to your hero typography. Humans catch that; AI doesn't.
Edge case decisions: What happens when a card title wraps to three lines on mobile? How does a disabled state look in dark mode? These micro-decisions define the character of your product.
Token Architecture in 2025
We've restructured our token system into three tiers: primitive tokens (raw values), semantic tokens (purpose-mapped), and component tokens (scoped to specific UI elements). This layered approach lets AI work safely at the component level while humans control the semantic and primitive layers.
Where This Is Heading
Design systems will evolve from component libraries into decision libraries. Instead of documenting what a button looks like, we'll document why it looks that way - the reasoning, the brand principles, the user research that informed the choice. AI can then use that reasoning to generate new components that are not just visually consistent but philosophically aligned.