Evidence-linked trend hypothesis
Design systems become executable agent context
Design systems are expanding from component libraries into agent-readable rules, references, registries, review loops, and constraints that preserve product character.
Current thesis
- The useful design system for an AI development workflow is an executable context package, not only a visual component catalog.
- Confidence: medium.
- Supported by 6 normalized raw signals.
Source ledger
Publishable sources attached to this record.
| # | Source | Role | Public status |
|---|---|---|---|
| 1 | youtube.comsource | primary receipt | source_urls |
| 2 | styles.refero.designsource | supporting receipt | source_urls |
| 3 | shadcn.runsource | supporting receipt | source_urls |
| 4 | stitch.withgoogle.comsource | supporting receipt | source_urls |
| 5 | github.comrepo | supporting receipt | source_urls |
What is becoming executable
A component name is not enough to reproduce a product’s design character. An agent also needs typography, spacing, color, layout constraints, preferred composition, anti-patterns, reference screens, and a way to inspect the result. ProtoDash, Refero, shadcn registries, Stitch skills, and Taste Lab package more of that environment into machine-usable form.
Supporting evidence
- ProtoDash combines Stripe’s design system with rules, components, sandboxes, and in-product review.
- Refero exports visual references as structured DESIGN.md files.
- shadcn registries make components and installation metadata discoverable.
- Stitch and Taste Lab preserve design intent as reusable agent context.
Operational consequence
Version design rules next to components, include representative states and counterexamples, and require browser screenshots at target viewports. The agent can assemble the interface, but rendered evidence remains the acceptance artifact.