Figma Specs: New Markdown Editor Sets Rules for AI Agents

Figma wants AI agents to follow a brand's design rules, not just produce something that looks plausible. At its first Config event in India this week, the design platform introduced a set of features built around one idea: give agents, both Figma's own and those from third parties, a clear and central reference for how a company designs.

The announcements cover a new Markdown-based editor, expanded context for design systems, a wider beta for code on the canvas, more tool integrations, and a tighter link with the Payload content management system that Figma acquired last year.

Figma Specs: a written definition of the design language

The headline feature is Figma Specs. It is a Markdown editor where teams describe their design language in plain, structured text. The resulting document works as a reference for human designers and for agents working inside Figma.

A Spec can live as its own file or sit directly on the canvas next to the designs it describes. It is more than a text box. Users can point to other Figma elements, such as items on the canvas, and embed FigJam diagrams, code layers and designs. FigJam is Figma's whiteboarding tool.

Teams can also pull in more context for their design system: design libraries, code libraries and written guidelines. Agents can then consult this central system when they generate or change work, so the rules sit in one place instead of being scattered across files and people's heads. The approach is similar in spirit to efforts to tie enterprise data to AI through a shared layer of context.

Steering a non-deterministic agent

The timing is not accidental. In the past week, Figma opened its on-canvas agent to a wider group of users. AI output is non-deterministic, meaning the same request can produce different results each time. Figma's answer is to provide tools that steer the agent toward more predictable outcomes, and Specs is the most direct example.

Code layers move to open beta

Figma also opened its code layers feature to everyone as an open beta. Code layers add code to designs on the canvas, which allows more interactive prototypes. The company first announced the feature in a private beta in June; the India event widens access so more users can test it.

Users can ask the agent to turn a design into code, or to change part of the code while they iterate on a prototype.

Loredana Crisan, Figma's chief design officer, explained the reasoning to TechCrunch. "We see Figma as the place where software teams come to decide what they build and ship," she said. She pointed to the gap between how much code is now written and how much actually ships: "There are 17-times more lines of code being produced today than in the past. But when you're looking at releases, only 1.3-times more releases. A lot of that code actually is not necessarily built to ship, but really built to discover what the product is. It's built to design."

In other words, Figma sees much of today's AI-generated code as a design tool in its own right, a way to explore an idea rather than a finished product.

More MCP tools and a Payload integration

Figma said its MCP support now covers more than 20 new tools. MCP, the Model Context Protocol, is a common way for AI tools to connect to outside software and data. The new additions include Open Code, Dia, Grok Bot and Sarvam, an Indian AI startup.

On the publishing side, Figma is connecting Payload to Figma Make, its product for generating websites. Payload is an open source headless CMS, a content management system that stores and serves content without dictating how the front end looks. The integration is meant to let users build dynamic websites and manage their content directly. According to Figma, Payload's downloads have grown sixfold over the last 12 months.

Why India

The choice of venue reflects Figma's priorities. The company opened its first office in India late last year, covering sales, partnerships, go-to-market and product support. Figma says India is its largest market by users outside the United States, and that it works with 40% of India's top 100 companies listed on the Bombay Stock Exchange, one of the country's two main stock exchanges.

Our Take

The most interesting part of this release is not the editor itself but what it implies: as agents take on more design work, the bottleneck shifts from generating output to keeping that output consistent. A plain-text definition of a design language, readable by people and agents alike, is a practical way to narrow the range of results without locking the agent into rigid flows.

This fits a wider pattern. Enterprise software vendors are trying to turn their products into places where agents operate under defined rules, as SAP is doing with Joule as an agentic work layer. Industry groups are writing shared rules for agents in other domains.

It is worth watching whether third-party agents connected through MCP actually respect Specs in practice, or whether the guidance mostly helps Figma's own agent. Crisan's code-versus-releases figures also suggest Figma is positioning itself upstream of engineering. Whether code layers and Figma Make pull real build work onto the canvas will become clearer once the open beta has run its course.