Figma's agent goes GA with design system context

Figma moved its agent out of beta into general availability on Oct 6 and published three posts about it in one day. Figma’s own measure of the change is that the agent now wins around 60% or more of its human graded evals with professional designers, and that it follows instructions better and handles longer running work than the beta build. That figure is Figma grading its own agent against its own evals, not an independent benchmark, so treat it as a direction of travel rather than a score to compare with other tools.

The GA release adds three capabilities beyond output quality. Library owners can upload markdown files to a Figma library so the agent learns how the team expects the design system to be used, with rules written in plain language, such as only using a particular colour for actions and never for decoration. The agent can find and insert context from across the platform, pulling in FigJam stickies, Figma Slides copy and Design frames. And users can follow collaborators’ agents as they work on the canvas, which Figma describes as a multiplayer model where several agents operate on the same file at once.

Figma names teams at Uber, Granola and Atlassian as the reference users, and the connectors run to Amplitude, Asana, Atlassian, Granola, Notion and Slack. Uber’s design systems team, which delivers components across seven platforms, used the agent with custom skills to produce development ready component documentation that “once took multiple people months” and now gets published by a single designer in an afternoon, with skills that map fills, strokes, shadows and gradients onto design tokens and flag hard coded values. Granola’s Paavan Buddhdev uses the Granola connector to pull meeting notes and transcripts onto the canvas as annotations, which he says buys back the attention he used to lose typing notes during crits.

(Original source: Figma)