Google Labs: Stitch's DESIGN.md format is now open source
Google Labs released the DESIGN.md specification as an open-source format in April 2026. The format originated in Stitch, Google’s AI-powered design tool, but the specification itself is now a shared standard that any team or tool can adopt, contribute to, or build on.
A DESIGN.md file captures a project’s visual rules in plain text: color roles, typography, spacing, component behavior, and written design intent. The purpose is to give AI coding agents something structured to read before generating UI, rather than having them default to generic styling patterns that have nothing to do with the actual design system. Tools like Cursor, Claude Code, Lovable, and similar environments can parse the file and use it to validate decisions — applying the right spacing tokens, selecting correct color roles, and checking generated output against WCAG accessibility rules.
The practical problem it addresses is well-known to teams that already work with AI coding tools. When an agent generates a screen or component without access to design system context, the result looks plausible but drifts from brand standards. Corrections happen after the fact, by hand. DESIGN.md moves that context earlier in the process, so agents generate with brand rules active rather than ignoring them.
The format supports portability. Designers can export design rules from Stitch and import them into another project, or author the file manually using the published specification. A GitHub repository maintained by the community already includes DESIGN.md files modeled on major brand design systems, which teams can drop into projects as a starting point without building from scratch.
Google Labs engineer Cassia Xu announced the release. The specification is open to contributions and maintained at github.com/google-labs-code/design.md.
Relevant for teams where designers and AI coding agents work from the same codebase. It resolves a specific friction: AI-generated UI that looks acceptable in isolation but diverges from brand standards once it lands in a production context.