Skip to content
Article Kaelig Deloumeau-Prigent Jul 2026

Kaelig Deloumeau-Prigent: state of AI in design systems, July 2026

Kaelig Deloumeau-Prigent, a design systems engineer, published this field survey based on data gathered July 26–28, 2026 from 20 open-source design systems and 6 design platforms. The research documents what these systems are shipping to make themselves readable and usable by AI coding agents — a concern that has become practical rather than theoretical as more teams rely on AI to generate UI components.

The survey covers two directions simultaneously. The first is consumption: how design systems ensure that coding agents build with real components rather than inventing their own or hallucinating variants. The second is contribution: how design system teams use AI internally to maintain the system itself.

Across the 20 design systems surveyed, the report identifies 187 affordances and 157 distinct “coercion” techniques — methods that steer agent behavior toward a system’s intended components rather than arbitrary alternatives. These range from publishing an llms.txt file, a machine-readable index of what is in the system and how to use it, to shipping an MCP server that lets AI assistants browse, search, and install components directly. The shadcn MCP server is cited as a concrete example: it gives AI tools access to the component registry, so a developer can ask an assistant to find and install a specific component without leaving their editor or doing a manual search.

The full report is queryable: it can be connected to an AI assistant so a design team can ask which techniques other systems have adopted and receive source-linked answers. This design — a living document that agents can read and traverse — is itself an illustration of the direction the field is moving.

For design system teams, the practical takeaway is that making components AI-consumable requires deliberate publishing decisions. A design system documented only for human readers is increasingly invisible to the coding agents that drive a growing share of front-end code generation. The survey gives teams a structured baseline for assessing their own AI-readiness against what leading open-source systems have already shipped.

The full report is also available at the zeroheight platform, which co-published it alongside the Netlify site. The GitHub repository at kaelig/state-of-ai-in-design-systems contains the raw data and source links for each technique.

Useful for design system leads, platform engineers, and anyone responsible for making a component library usable in an AI-assisted development environment.