Figma introduces on-canvas AI agent

Figma has moved its AI push from side-panel assistance into the design surface itself. In a March 24 blog post, the company said users can now use AI agents to “design directly on the Figma canvas,” generating initial concepts, editing existing work and producing variations inside a file rather than outside it. The feature is tied to what Figma calls “skills,” or team-specific instructions that give agents context about components, naming and design decisions. A May 20 post in Figma’s forum said the company’s AI agent was entering closed beta. That post described the tool as built directly into the Figma Design canvas and said it works with a team’s design system, components and context. It listed tasks including generating multiple layout options, trying style variations and converting screens to dark mode. ### So what exactly did Figma introduce? (figma.com) Figma described the new capability as an AI agent with write access to the canvas. In its March 24 announcement, the company said agents can create and modify designs directly in Figma files, rather than only suggesting text prompts or external mockups. The post said users can guide those agents with “skills,” which are reusable instructions that capture team conventions and intent. (forum.figma.com) TechCrunch reported on May 20 that Figma’s agent can be directed with natural-language prompts to generate new designs, edit existing ones and automate tasks such as producing iterations of current designs. The report also said users can run multiple agents at the same time. ### How is this different from earlier Figma AI tools? (figma.com) Figma has already been rolling out a broader set of AI products under the Figma AI and Figma Make labels. An April 2 company post said “Make kits” and “Make attachments” were added to bring real components, data and constraints into Figma Make prototypes. A separate product page describes Figma Make as a tool for turning ideas into interactive, testable prototypes. (techcrunch.com) The new agent feature is narrower and more direct: it operates on the canvas itself. Figma’s Help Center says the “code to canvas” workflow shown in its May 2026 release-notes livestream uses agentic tools to start from either production code or the Figma canvas and then write results back into design files. ### What can the agent do inside a file? (figma.com) Figma’s forum post gave the clearest task list for the current beta. The company said users can ask the agent to explore more directions through layout options, test style variations and convert screens into dark mode. The March 24 blog post also said agents can help with design work while following team-specific guidance supplied through skills. (help.figma.com) Figma said those skills can encode decisions about which components to use and how teams want work structured. The company said the feature is free during beta and is expected to become a usage-based paid product later. ### Where did the May 20 screenshots fit in? A May 20 X post cited in the source briefing circulated screenshots and a short description of the on-canvas agent. (forum.figma.com) That social post matched Figma’s official description of a tool that can generate initial UI concepts, propose variations and iterate on designs from within the canvas. The official materials, however, show the feature itself was first announced on March 24, with a closed-beta update posted on May 20. (figma.com) ### What happens next? Figma’s Help Center points users to workflow examples from the May 2026 release-notes livestream, and the company’s forum says the on-canvas AI agent entered closed beta on May 20. Figma’s March 24 blog post said the feature is free during beta and will eventually shift to usage-based pricing. (help.figma.com) (figma.com)