Generative UI

Key takeaways

• Generative UI lets AI assemble interfaces at runtime from user context and tools, with the design system as the guardrail that keeps it coherent.
• Fix the components, tokens, accessibility, layout templates, and validation schemas; leave only data and flow branching flexible.
• The runtime flow validates each declaration in the renderer and renders only with approved design-system primitives.
• Control key risks by rendering only approved components, requiring action metadata and confirmation, and logging declarations for debugging.
• Good fits are data exploration, ops dashboards, and copilots; avoid generative UI for legal, financial, or safety-critical flows unless tightly governed.

Traditional UI is defined ahead of time by developers. Generative UI lets AI assemble interfaces at runtime based on user context, task state, and available tools. The design system is the guardrail that keeps this flexible UI coherent.

What Must Be Fixed

Generative UI should not mean arbitrary UI.

Runtime Flow

1. User asks for a task.
2. Agent determines intent and available data.
3. Agent selects approved UI components.
4. Renderer validates the declaration.
5. UI renders with design-system primitives.
6. User actions feed back into the agent or backend.

Risks

Good Use Cases

• Data exploration.
• Support and operations dashboards.
• AI copilots that assemble task-specific controls.
• Internal tools with many low-volume workflows.

Avoid generative UI for legal, financial, or safety-critical flows unless the renderer and action policy are tightly governed.