DESIGN.md for generated UI
- A new video introduced DESIGN.md, an open specification for AI‑generated user interfaces and design artifacts. - The spec proposes a machine‑readable pipeline that converts intent into UI components and supports human review. - A shared intermediate format could let teams version, render, and validate generated interfaces across frameworks more reliably. (youtube.com)
Design systems are the rulebooks for colors, type, spacing, and components that keep an app from changing its look on every screen. On April 22, Google published DESIGN.md as an open-source draft spec for turning those rules into a plain-text file that coding agents can read. (github.com) The project’s GitHub repository describes DESIGN.md as “a format specification for describing a visual identity to coding agents,” and says it gives agents a “persistent, structured understanding” of a design system. A Google for Developers video published the same day said the format began inside Stitch and is now being opened up beyond a single tool. (github.com) (youtube.com) In practice, the file has two layers. The spec says a DESIGN.md document can include YAML front matter for machine-readable tokens like colors, typography, spacing, rounded corners, and components, followed by markdown prose that explains the reasoning behind those choices for humans and models. (github.com) Google’s video frames that split as the “why” plus the exact values. The machine side carries details like hex codes and font sizes, while the prose side tells an agent how to use those values as semantic roles instead of treating them as random variables. (youtube.com) (github.com) That pitch lands at a moment when Google is expanding Stitch, its Google Labs design product, into what it called an “AI-native software design canvas” on March 18. In that update, Google said Stitch can generate high-fidelity user interface designs from natural language and import or export design rules through DESIGN.md. (blog.google) The underlying problem is simple: most design systems live in tools and documents that humans can read, but agents cannot reliably follow. Google’s spec says DESIGN.md is meant to serve as a “living source of truth” that humans and AI can both understand and refine across design sessions and tools. (github.com) The format is also designed to travel. The spec says its token structure is inspired by the Design Tokens community format and that the values are meant to convert to or from tokens.json, Figma variables, and Tailwind theme configs. (github.com) Google is also pairing the format with tooling, not just a document. The video says a new command-line linter can validate design updates against Web Content Accessibility Guidelines, or WCAG, before an agent writes code. (youtube.com) That makes DESIGN.md less like a mood board and more like a shared intermediate file: one versioned in Git, reviewed in markdown, and checked by software before it turns into interface code. The opening move is modest — a draft spec and a short explainer video — but the goal is to make generated user interfaces easier to reproduce, inspect, and keep on-brand across tools. (github.com) (youtube.com)
