Figma launches Specs, a Markdown home for agent context, at Config India

What did Figma announce at Config India?

Figma used Config India, which it calls its largest international conference yet, to publish a batch of launches on 15 October 2026. The umbrella post, “Build by design: New workflows for new ways of building” by co-founder and CEO Dylan Field, groups them around one argument: AI has made code cheap, so teams now build quick prototypes “to think” and then separately build carefully “to ship,” and Figma wants both kinds of work, and the agents doing some of it, to happen on its canvas.

Five things were announced. Specs is a new home for Markdown context. Code layers, which Figma first showed earlier this year, moved into open beta. Design systems gained guidelines and code sources. Payload, the open-source content framework whose team joined Figma last year, is now built into Figma Make. And the Figma MCP server gained new capabilities and 20 new partner integrations. Separately, Figma announced an MCP partnership with the Indian AI company Sarvam and a student programme in India.

Field framed the problem with a customer quote: building with AI is “fast, fast, fast... then slow,” because vibe-coded prototypes are usually single-player and the reasoning behind them disappears into chat logs and terminals. The launches are Figma’s answer to where that context and code should live.

What is Figma Specs?

Specs is a container for written context that lives on the Figma canvas. The launch post, written by product manager Luke Zhang, tech lead Johnny Dallas and designer Aosheng, starts from a problem many teams working with agents will recognise: Markdown has become the default format for sharing context with agents, but Markdown files tend to sit on one person’s computer, with brittle links to each other and little connection to the project they describe.

A spec fixes some of that by being multiplayer. You can drag an existing .md file onto the canvas, press D, or click the Specs icon in Figma or FigJam to start one. Inside a spec you can embed a FigJam file, a frame or any node from the canvas, so text and designs sit side by side rather than in screenshots or stale links. Teammates can duplicate a spec so that each person works on their own version before the team aligns on a direction, and elements can be dragged out of a spec, edited, and dragged back in while keeping their Figma properties.

The Figma agent, which became generally available earlier in October, can read and reply to comments in a spec. Figma suggests asking it to answer every open comment with a suggested fix while people weigh in on the same document, with each conversation in its own thread. Connectors pull in outside context: a GitHub connector can draft a technical spec from a codebase, and a Hex connector can run a data analysis and leave the result as a comment beside the question it answers.

How do Specs reach coding agents?

The hand-off is the point. Once a team agrees on a direction, Figma says you can copy the link to a spec and paste it into a prompt for your coding agent. Through the Figma MCP server, an agent such as Codex can then read the spec as Markdown along with what it links to, including references to frames, diagrams, code layers and Code Connect snippets. If the protocol is new to you, our explainer on the Model Context Protocol covers how a server like Figma’s exposes tools and context to an agent.

Figma gives an example from its own team. Software engineer Noah Finer wrote a spec of guiding principles and edge cases while building the Specs feature itself, embedded FigJam diagrams of the flow and collected comments. He then gave the finished spec and his coworkers’ comments to his agent via the MCP server, used MCP to update the spec as the code progressed, and had agents create new FigJam diagrams for the engineers reviewing his code.

Specs also works in the other direction. The Figma agent can turn a product requirements document into designs, prototypes and code, or turn designs and code back into a spec. For a code layer, an Extract designs command converts its current state into editable design layers, so engineers can mark frames ready for development and keep them next to the spec that describes them.

Code layers: prototypes backed by real code on the canvas

Code layers bring working code onto the shared canvas. In the code layers post, Figma product designer Sean Blanton says teams can convert existing designs into working code, import a supported lightweight prototyping codebase from GitHub or a local folder, bring a coded prototype in from another tool through the MCP server, or capture a live web page with Figma’s Chrome extension. A code layer imported from GitHub stays connected to that repository.

Once on the canvas, a prototype behaves like other Figma objects. Teammates can duplicate it to try an alternative and compare versions side by side, comment on it, annotate frames to show the agent where a change is needed, or edit the code directly. The Figma agent can add features or combine the layout of one version with the interactions of another. When a change is ready, you can push the code to a branch and open a pull request so the team reviews it before merging.

Figma quotes several customers. It says the agency Merkle reports 30% faster iteration and feature development and 50% less documentation such as interaction specs on tickets, and a Checkout.com designer says code layers helped him build a client demo about 50% faster. Those are customer figures published by Figma, not independent measurements. The company also says it plans to support more complex codebases over time, which implies the current beta is aimed at prototypes rather than production apps.

Design systems, Payload and 20 new MCP integrations

For design systems, Figma is adding guidelines, written rules attached to published libraries so the agent can follow established patterns, and code sources, which import components from a GitHub repository or an npm package so the agent builds better components and code layers. It is also previewing a central place to manage a design system’s libraries, with a system map that tracks changes and dependencies and a system check that finds gaps such as missing descriptions or interactive states.

Payload now runs inside Figma Make. According to the Payload post, prompting an app in Make can also set up a database, file storage, user login and an admin panel, which Figma hosts. Marketers can edit content at the site’s /admin path, developers can customise the app with Payload’s open-source framework, and API keys can be stored securely without a separate backend. Figma says Payload downloads grew 6x in the past 12 months and describes it as the fastest-growing open-source CMS on npm. Google has made a similar bet that non-developers will ship what they prompt, letting people run private vibe-coded Android apps from AI Studio.

The MCP server got broader too. From today, Figma says, AI tools connected to it can generate code layers, work with specs, create visual assets and run published Weave workflows, and better agent guidance and server performance reduce errors and token usage. The 20 new MCP partner integrations include OpenCode, the Dia browser, JetBrains’ Junie and Meta Muse.

Why India, and what Figma said about the market

The venue was chosen for a reason. In its India announcement, Figma says India was its second-largest market by daily active users globally as of 30 June 2026 and one of its fastest-growing, and that more than 30,000 people registered for Config India in person and virtually. It says more than 40% of the BSE100 companies use Figma, naming Airtel, CARS24, Groww, Juspay, Myntra, Swiggy, TCS and Zomato.

Sarvam’s Indus is joining Figma’s MCP ecosystem alongside Claude Code, ChatGPT, Cursor, GitHub Copilot, Xcode and VS Code, in what Sarvam CEO Pratyush Kumar called an India-first AI collaboration. Figma is also expanding its Campus Leaders programme to Indian universities, its first cohort outside the US and Canada. Applications close at 11:59 p.m. IST on 7 November 2026 and the cohort starts in January 2027.

Figma also published survey results: 507 responses from Indian designers, developers and product managers, part of more than 1,800 surveyed worldwide between 17 August and 16 September 2026. It reports that 65% of Indian respondents are especially excited about how work is changing, against 44% in the US and EMEA, and that 83% say they are more productive. This is vendor research about Figma’s own users and prospects.

Who gets what, and what it costs

Availability differs by feature, and the posts are specific about it. Specs is rolling out in open beta on web, desktop and mobile for paid plans, starting today. Code layers is rolling out in open beta on all plans and uses AI credits, but creating and editing code layers is free on paid plans through 4 November 2026. Design-system guidelines are in open beta and code sources in closed beta, both for paid plans. Payload in Figma is available on all plans, with access rolling out to all accounts over the coming weeks.

Figma did not publish new prices in these posts, and it does not say how many AI credits a typical code layer session uses once the free period ends. Teams planning to lean on code layers after 4 November should check the AI credit allowance on their plan in Figma’s help centre before building workflows around it.

What to watch before relying on it

These are betas, and every productivity figure comes from Figma or its customers. We have not tested Specs or code layers. The bigger open question is whether a shared spec actually keeps agent-built work aligned, or just adds another document to maintain. Research on coding agents suggests that writing code is no longer the bottleneck: a Harvard study of 718 firms found agents add more code without more finished work, because review becomes the constraint.

Specs and code layers are aimed at that gap, putting intent, prototypes and review in one shared place. Whether that works will depend on how well agents follow a spec in practice, how code layers handle real codebases beyond lightweight prototypes, and what the AI credit costs look like after the free window. Those are the things to watch as the betas roll out to more accounts.