The Problem
Design systems teams have operated as component factories for over a decade. We designed components in Figma, built them in React, wrote documentation, and spent most of our energy on adoption — office hours, migration guides, scorecards, evangelism.
That model was built on an assumption that's no longer true: that the distance between design intent and production code was large, expensive, and required dedicated human effort to bridge.
AI-assisted development has collapsed that distance. Engineers working with tools like Cursor and Claude Code can generate compliant UI in minutes. Designers are increasingly working in code-aware environments, shipping PRs and vibecoding applications. The bottleneck is no longer "how do we get teams to use our components." It's how do we ensure that everything being generated — by humans and AI agents alike — is structurally sound, accessible, and coherent across products.
The design system's job has shifted from building the thing to defining the rules that govern how things get built.
Three Core Beliefs
1. The system's most important consumer is now an AI agent.
When an engineer prompts their IDE to "build a settings page," the design system should already be in the room — loaded as context, shaping output before any human reviews it. Deliverables must be machine-readable specifications, not just human-readable documentation. .cursorrules files, CLAUDE.md context documents, MCP server configurations, and structured schemas are now first-class system artifacts — as important as Figma libraries, possibly more so.
2. Adoption is an infrastructure problem, not a communication problem.
The old model spent enormous energy convincing teams to adopt the system. In the new model, adoption happens by default because the system is embedded in the toolchain. When AI-assisted generation already knows your tokens, component APIs, and composition patterns, the path of least resistance is the system. We stop treating adoption as a sales motion and start treating it as a platform engineering concern.
3. Governance scales through automation, not process.
Contribution committees and intake request forms are artifacts of a world where compliance could only be checked by humans at low frequency. Compliance checking should be continuous and automated — evaluating whether new UI adheres to system patterns in real-time. The job isn't to review every contribution. It's to build the validation layer that makes non-compliance visible immediately and resolution frictionless.
What needs to change
The team becomes more technical. The people who matter most now operate at the intersection of design system architecture and platform engineering — defining machine-readable constraint layers, owning token pipelines and compliance automation, and building the AI integration infrastructure that makes the system self-enforcing.
The source of truth is what ships. Static Figma files and pages of documentation were never really the source of truth. Production is. The design system's job is ensuring that what ships is correct, through specifications — machine-readable constraints expressed in the tools teams actually use. Code is an artifact of those specifications. Figma remains a design environment, not the canonical reference.
Creative authorship doesn't go away — it gets amplified. The design system team doesn't just codify and distribute. It defines how the brand expresses itself in product — motion, interaction personality, the editorial judgment about where and how much expression is appropriate. Infrastructure speed doesn't replace the need for that authorship. It makes those decisions travel further and land more consistently.
What Doesn't Change
Design quality still requires human judgment. AI can generate compliant UI but can't evaluate whether a flow is coherent or whether an interaction model serves the user's actual mental model.
Accessibility is non-negotiable — and it's evolving. LLM-powered assistive technology will shift what "accessible" means, but we're not there yet. The team still needs deep accessibility expertise, increasingly expressed through the tooling and verification layers rather than manual review.
The Measure
This is working when a new designer or engineer can prompt their AI tools and get system-compliant UI without ever reading the documentation. When system drift is caught in automation, not audits. When the team's architectural influence grows even as its headcount stays lean. When more time goes to the hard design problems — coherence, accessibility, brand expression — and less to the mechanical ones.