Figma’s Agent Is Turning the Design System Into Software

DesignWhine, a design publication, argues that the AI agent in the design tool Figma can now apply a design system's written rules, so unclear rules will spread mistakes faster.

DesignWhine's editorial team writes about the AI agent in Figma, a design tool, which became generally available on 6 October 2026. Teams can attach written guidelines to a design library, and the agent can update components in bulk, refresh documentation and review proposed changes before they are published. The authors argue that a design system, the shared components and rules a product team builds from, is changing from a reference people read into rules software applies.

• An AI agent cannot enforce a rule that nobody wrote down, and cannot tell the current component from an old one when both look valid.
• Coinbase, a cryptocurrency exchange, linked its Figma components to its production code. In three test runs, an AI coding agent's implementation time fell by 22.3% and its cost by 22.5%, on Coinbase's figures.
• The authors expect design-system teams to work like internal platform teams that version the agent's instructions, maintain the links to code and check the agent's output.
• They list eight statements for testing a design system, including whether the current component is obvious and whether an AI agent's bulk changes can be reviewed before publishing.

The authors argue that a team should make its design-system rules explicit before letting an AI agent maintain the library, because the agent repeats whatever ambiguity it finds. The piece is commentary on a vendor's product, and its numbers come from case studies that Figma published.

Derived by Working Surface from the article; more in the Playbook. Source line: An agent cannot enforce a rule that the system has never made explicit.

10 October 2026: as AI agents start work without being asked, the rules that limit them need to be written down and owned: by a person, a team and the machine itself.