Figma Adds a Native AI Design Agent to Its Canvas

Figma began rolling out its first native AI design agent on 20 May, embedding a chatbot-style interface directly inside the collaborative canvas its users already work in. The agent accepts plain English instructions to generate designs, edit existing layouts, and produce iteration sets, and it can run as multiple parallel agents handling different parts of a file at once.

The product positions Figma alongside a growing crop of "agentic" design tools, but with the home-field advantage of already owning the canvas. Until this week, users who wanted AI assistance in Figma generally had to bolt on third-party plug-ins or move work into separate tools. Now the model sits inside the file itself, with access to layers, components and the project's design system, which the company says lets the agent stay consistent with house style rather than producing context-free output.

Built on partner models

The launch arrives on the back of partnerships with Anthropic and OpenAI, and Figma has confirmed support for AI command-line tools like Claude Code and Codex to run alongside the design environment, as TechCrunch reported on launch day. According to coverage by Dataconomy, the agent itself is built on a stack of fine-tuned models from both labs, with Figma routing tasks to whichever model performs best for a given design operation.

Rollout is gradual. The agent is available to eligible Figma Design users from 20 May and will expand in waves over the coming weeks. During the beta period, agent usage will not consume AI credits; once the feature reaches general availability, credit consumption applies as it does for Figma's other AI features.

The real test is the boring middle

For working designers, the question is less whether the agent is impressive than how it handles the boring middle of the job: making forty variations of a hero section, applying a colour change across an entire prototype, or generating mobile counterparts to a desktop layout. Those are the tasks that consume hours and that previous Figma AI features handled unevenly. If the agent gets them right, it changes the daily rhythm of design work in the same way coding agents have changed the daily rhythm of writing software. If it doesn't, it becomes another impressive demo that quietly drops out of the workflow.