Saturday, August 1
Saturday, August 1
Saturday, August 1
Medium · BootcampCristina Stoica
AI summary focused on the core points of the original article.
Modernizing enterprise UX requires more than introducing a design system—such as a UI kit or component library—or adding AI. True UX modernization must begin with context engineering: explicitly encoding implicit business logic and workflow architecture.
Older systems are not merely liabilities to discard. They contain implicit knowledge accumulated over years, including business rules, compliance requirements, and exception-handling cases.
Modernization should not remove complexity indiscriminately. It should distinguish essential domain logic from unnecessary operational friction and redesign each accordingly.
Design systems define “nouns,” such as buttons, modals, and tables, but they do not provide the “verbs”: interaction rules such as role-based navigation, recovery paths after network errors, and conditions for two-factor authentication.
Without this system context, AI tools such as Cursor and Claude must rely on assumptions, producing inconsistent results.
Design intent should be maintained alongside the codebase in a machine-readable document such as DESIGN.md.
Machine-readable tokens: Style values including spacing, color, and typography
Behavioral heuristics: State-transition rules and layout responses to exceptions
A component library documents what has been built, while DESIGN.md documents how the system behaves.
Producing high-quality UI without first establishing the underlying logic is one of the most common mistakes in modern software design.
Design leadership is not limited to pixel-level execution. Its role is to translate complex domain constraints into machine-readable guidance that both human developers and AI agents can understand.
This article is useful for understanding what needs to be documented after a design system is established, and how that context directly affects the quality of work produced by AI agents.
Good time to read
When a design system exists but product behavior and exception handling remain inconsistent
When you want to reduce the amount of design intent an AI agent has to guess inside an enterprise codebase
Key takeaways
A way to interpret legacy workflows as sources of domain knowledge rather than removal targets
How to use DESIGN.md as a shared system-behavior contract for humans and AI rather than a style document