Disclosure: the author runs Superprompt Pro, a product design practice that builds brand and design systems, and Ship Index Grow, which teaches the same workflow to product teams.
Direct answer: Figma has opened its canvas to agents. Skills files turn a design system from a reference document into instructions an agent reads before it edits anything.
Figma has opened its canvas to agents. Through an MCP client such as Claude Code or Codex, an agent can create and change design assets that stay linked to a team's design system. The feature is free during beta and is expected to become a usage-based paid feature later.
The most interesting part is not the canvas. It is the skills layer. Skills are Markdown files that tell an agent which steps to follow and which conventions to apply. In practice, they turn a design system from a reference document into a set of instructions an agent reads before it touches anything.
That changes what a design system is for. Three things follow:
For a team that already has a design system, the next step is an audit: which rules exist only in a PDF, and which ones are written so an agent could follow them? The gap between those two is the work.
Why a skill is not the same as a guideline
Most design systems were written for people. They explain intent, show examples and expect a reader to fill the gaps with judgment and context. Agents can read them too, but they cannot always tell which sentence is a rule and which is advice. A skill file narrows that gap by stating steps and conventions in the order an agent should apply them. The difference is subtle but practical: a guideline asks for good judgment, while a skill asks for a procedure that can be checked.
A checklist for writing agent-readable rules
Several tests make a rule easier for an agent to follow. A rule should name the artifact it governs, such as a component, a token or a layout. It should say what to do and what not to do, with the forbidden case stated plainly. It should point to the variable or component that replaces the forbidden value, so the agent has an alternative and does not improvise. And it should say how to verify the result, for example by comparing against a named component or checking auto layout settings. Rules that fail these tests can still work when a senior designer applies them, but they will produce inconsistent output from an agent.
The design system team gets a new job
If agents edit real components, the design system team stops being only a library maintainer. It becomes the team that decides which changes are safe to automate and which still need a human decision. Components that many screens depend on deserve stricter rules and clearer ownership. Tokens that encode brand decisions should be locked more tightly than tokens that encode layout detail. Those decisions are product decisions, and they should be made deliberately before automation reaches the parts of the system that are hardest to undo.
What to do this quarter
For a team with an existing system, the practical first step is an audit of its rules. Mark each rule as either written for people only or written so an agent could follow it. Then convert the most-used rules first, starting with spacing, color and typography, because agents touch these most often. Review a sample of agent output against the converted rules every week for the first month. The gaps in that sample are the most accurate measure of how ready the system is for automation.
The same discipline applies to the agents themselves. Each rule you convert should come with an example of a correct output and an example of a rejected one. Those pairs become the test set for the next version of the rules, and they reveal quickly where a rule is ambiguous. Teams that keep these pairs find that most arguments about the system are arguments about two or three rules, and fixing those rules settles the rest.
Related reading
Sources
- "Agents, meet the Figma canvas" (Figma blog, publication date not stated on the page): https://www.figma.com/blog/the-figma-canvas-is-now-open-to-agents/