Figma defines how agents work with Specs and code layers

Seventeen times more lines of code, but only 1.3 times more releases. That was the ratio Loredana Crisan, Figma's Chief Design Officer, used to frame the set of features the company presented this week at its Config event in India, as reported by TechCrunch. The diagnosis is clear: agents are generating code faster than teams can turn it into shipped product, and Figma wants to position itself right at that bottleneck.

Every announcement revolves around the same idea: giving AI agents an explicit framework to work inside the platform, instead of letting them infer design rules from screenshots or scattered files.

Figma Specs: the design system in markdown

The centrepiece is Figma Specs, a markdown-based editor for defining a product's design language. It can work as a standalone file or live inside the canvas, and it lets you reference Figma elements and embed FigJam diagrams, code layers and designs. The stated goal is twofold: to serve as a reference for the human team and for the agents working on that same project.

Anyone using Claude Code will recognise the logic. A well-written CLAUDE.md or skill completely changes the quality of what the agent produces, because it provides context the agent cannot infer on its own. Specs brings that idea to design: instead of the model guessing spacing, tokens or tone of voice by looking at screens, it receives a structured, machine-readable document. Choosing markdown is no accident; it is what LLMs digest best today.

Code layers moves to open beta

The second announcement is the public beta of code layers, the feature that adds code directly to designs on the canvas to build interactive prototypes. Figma had announced it in private beta in June 2026. With it, agents can convert designs into code or modify existing sections of code without leaving the file.

It is one more step in the direction Figma has been taking all year: blurring the line between the design file and the repository. For small teams that prototype quickly, having real interaction inside the canvas saves a round of translation between design and development.

More agents connected through MCP

In parallel, Figma has opened its on-canvas agent to a wider user base over the past week, and its MCP support adds more than 20 new compatible tools, including Open Code, Dia and the Indian startup Sarvam. It confirms something we see every day: MCP has become the standard route for third-party agents to read and write in design tools, and Figma would rather be present in all of those clients than force people to use its own.

Figma Make, its app generation product, now also integrates with Payload, the open-source headless CMS, to build dynamic websites and manage their content directly. According to TechCrunch, Payload downloads have grown sixfold over the past twelve months.

Why India

The setting makes sense. Figma opened its first office in India in late 2025, the country is now its largest market outside the United States, and the company works with 40% of the top 100 companies listed on the BSE. Including Sarvam among the new MCP integrations fits that local bet.

Who it is useful for

For product teams with a mature design system, Specs could be the way to write down rules that today live in the heads of two or three people, and to get agents to respect them. For studios and agencies that deliver prototypes, code layers in open beta lets them try design-to-code workflows right away without requesting access. And for developers working with Claude Code or other MCP clients, the expanded support means less friction when asking an agent to read a frame and implement it.

At ElephantPink we use the Figma MCP every day, and our impression is that the problem was never that the agent could not code, but that it did not know the project's rules. If Specs manages to keep those rules in a single readable place, it will be the announcement that shows most in the final result; we will need to check that once it moves from demo to real work.