The shift
AI has made UI generation essentially free. A product team can prompt an agent and get a working, rendered screen in minutes — layout, components, interactions, responsive breakpoints, the works.
That’s not a threat. That’s leverage.
The question for design system and product engineering teams isn’t how do we survive generation? It’s how do we use generation to multiply our craft?
What generation can’t do on its own
AI generates. It doesn’t curate. It produces volume without inherent judgment about what makes output good — not just functionally, but experientially.
When I first tested AI against our component libraries back in 2024, it surprised me with what it got right. But it also added superfluous ARIA attributes that conflicted with our Angular directives. It didn’t know our components already had accessibility baked in. It couldn’t render, test, or feel what it produced.
The models are significantly better now. But the gap remains: AI can generate from your craft standards, but it cannot generate your specific discretion, taste and judgement, unless you are involved.
That’s your team’s job. And in an era of high-volume generation, it’s a bigger, more strategic job than ever.
Your value proposition, sharpened
The answer isn’t to slow down generation. It’s to ensure your craft is what generation draws from — and to keep raising that bar. Every layer of your design system is a value proposition worth sharpening.
Visual design. Your color system, typography, and spatial scale aren’t just aesthetic decisions. They’re the visual identity of your product experience. Design tokens make these decisions machine-consumable. The more precise and semantic your token architecture, the more coherent everything generated from them becomes. Invest in token naming conventions that encode intent, not just value.
Interaction design. Component behaviors — hover states, focus rings, error patterns, loading states, transitions — are where brand character lives. AI can copy the shape of an interaction pattern; it can’t originate one with intention. Define your patterns explicitly. Document the why, not just the what.
Motion design. Animation is underinvested in most design systems, but it’s one of the sharpest differentiators. Easing curves, duration scales, choreography patterns — these communicate quality at a felt level. As AI generates more UI, motion becomes one of the clearest signals of craft versus commodity.
Token architecture. Semantic tokens — surface-warning, text-action-primary, border-focus — are the vocabulary AI needs to generate meaningfully. Raw hex values and pixel numbers collapse under volume. A well-structured, well-documented token system becomes a multiplier for everything generated on top of it.
Technical offerings. Your component APIs are the interface between your craft decisions and every developer (and AI agent) consuming them. Predictable props, clean composition patterns, and thoughtful defaults reduce misuse at scale. The quality of your API design directly determines the quality of what gets built with it.
Compositional architecture — where components are composed together rather than controlled through an ever-growing prop surface — is essential here. Prop-driven approaches that try to handle every variant through configuration become unscalable fast; we’ve lived that lesson. Composability is the answer for productivity and longevity. Beyond the libraries themselves, all your component and token metadata should be surfaced on your documentation site in structured, machine-readable form. Add an llms.txt to your site and include LLM-optimized copy on your docs pages — concise, precise, contextually rich — so that AI tools consuming your system get accurate, high-quality context by default.
API integrations and ecosystem. The next frontier for design systems is becoming natively consumable by AI agents and coding tools. Offering your design system as an MCP (Model Context Protocol) server makes your components, tokens, patterns, and guidelines directly accessible to any coding agent that supports MCP — turning your system into live context rather than static documentation. Alongside that, building Skills for coding agents — curated, tested prompt packages that teach agents how to use your system correctly — ensures generated output respects your conventions from the start. That’s where craft gets enforced structurally, not just aspirationally.
Build craft into the generation pipeline
This is the lever most teams haven’t fully pulled yet.
If AI agents are generating UI from your design system’s assets, your assets become the curriculum. The quality of your documentation, metadata, and structured specs directly impacts the quality of everything generated downstream.
A few principles we’re acting on:
- Encode more of your craft into structured, machine-readable formats. Component metadata that includes composition rules, anti-patterns, and accessibility requirements — not just prop types.
- Build automated guardrails. Token compliance checks. Accessibility audits that understand your specific library’s built-in behaviors. API pattern validation. Not to slow things down, but to make generation safe at speed.
- Treat your docs as a first-class product. If your documentation is vague, generated output will be vague. If it’s precise, intentional, and well-structured, it trains better outcomes.
Craft in, craft out.
The human layer stays essential
Automation catches what it’s programmed to catch. It doesn’t catch what feels wrong.
The judgment calls — this is technically correct but experientially off — remain deeply human. The years of knowing why the system works the way it does, not just what the rules are, can’t be automated.
This is why our community touchpoints matter more in an age of high-volume generation, not less. In our weekly office hours across North America, Asia Pacific, and EU, we see AI-generated code that passes every automated check but misses the spirit of the system. That’s where craft gets transmitted. Where judgment gets calibrated.
AI can check the rules. Humans understand the reasons behind them.
The opportunity
When everyone can generate, the differentiator is quality. Intentionality. The felt sense that a product was made with care.
Design system teams have always been in the business of scaling quality. The mission hasn’t changed. The scale has — and so has the leverage.
Generation is a superpower. Craft is what makes it yours.
Keep tending to it, my friends. 😉