Figma Gives AI Agents a Markdown Rulebook for Design

Figma used its first Config event in India to make design systems machine-readable. The company unveiled Figma Specs, a Markdown-based editor that lets a team write down its design language once and then hand the same document to human designers and to AI agents working on the canvas.

A rulebook for non-deterministic agents

Figma opened its on-canvas agent to a wider user base in the past week, but generative models do not follow brand rules by default. Specs is the answer: a plain-text specification that can live as a standalone file or sit directly on the canvas. Authors can reference other Figma elements, embed FigJam diagrams, code layers and designs, and pull in design libraries, code libraries and the new Guidelines system so that context stays in one place. Agents read from that single source instead of improvising.

Code layers open up, plus more MCP tools

The bigger shift for builders is that code layers moved from private beta into open beta. A code layer holds real, running code, so a design becomes an interactive prototype on the canvas. Teams can ask the agent to turn a design into code, or edit part of that code to iterate on a prototype. Figma also expanded its MCP surface to more than 20 tools, including Open Code, Dia, Grok Bot and Indian startup Sarvam, and wired Payload, the headless CMS it acquired last year, into Figma Make for dynamic sites. According to Figma's own release notes, Specs, Code layers and Guidelines are all in open beta, while centralized code sources in libraries remain in closed beta.

Why it matters for builders

Figma's chief design officer Loredana Crisan framed the moment bluntly: 17 times more lines of code are being produced today, but only 1.3 times more releases. Most of that output is exploration, not shipping. When agents generate that volume, the bottleneck stops being generation and becomes instruction: which rules apply, which components exist, which context is authoritative. Specs is that instruction layer rendered as text, the same pattern now spreading through agent stacks, where a versioned rules file replaces a prompt.

For teams building agents against design or content systems, the practical lesson is to treat brand and design rules as structured, citable context rather than something an agent should infer. Figma is not the only vendor moving this way, but tying the rulebook to the canvas, the code layer and an MCP endpoint in one product makes the pattern concrete: the interface agents need is written down, versioned and shared.

Figma did not announce pricing or a general availability date for Specs.