Figma used its first Config conference in India, held in Bengaluru on October 15, 2026, to ship a set of features aimed at one problem: keeping AI agents and human teams working from the same design intent. The headline additions are Specs, a Markdown-based document format that lives on the Figma canvas, and the move of code layers, which embed working code inside design files, from limited testing into open beta.
Specs is meant to hold written context such as product requirements, technical plans and design-system rules. A spec can stand alone as a file or sit beside designs on the canvas, and it can embed FigJam boards, frames and other canvas elements so text and visuals stay connected instead of drifting into screenshots. Dropping an existing .md file onto the canvas converts it. Figma's agent can read a spec and answer comments inside it, connectors such as GitHub and Hex can pull outside context in, and a finished spec can be handed to an external coding agent through Figma's MCP server. Paid plans are getting Specs as an open beta across the web, desktop and mobile apps.
Code layers, introduced in a private beta in June, lets teams build and compare prototypes backed by real code without leaving the design canvas. The code can come from converting existing designs, a local folder, a GitHub repository, a coding agent connected over MCP, or a live web page captured with Figma's Chrome extension, and a code layer can be turned back into editable design layers. Figma says the open beta is available on all plans and draws on AI credits, with creation and editing free on paid plans until November 4, 2026.
Design systems get a parallel update aimed at agents as much as people. Teams can attach written guidelines to published libraries, now in open beta, and pull in code sources, either a GitHub repo or an npm package, starting in closed beta, so the agent has a fuller picture when it generates components. Figma also previewed a central place to manage libraries, with a map of dependencies and a check that flags gaps such as missing descriptions. Separately, Payload, the open-source content framework whose team joined Figma last year, is being built into Figma Make so teams can ship data-backed websites with content hosted by Figma; access is rolling out across all plans over the coming weeks.
On integrations, Figma said its MCP server can now create code layers, work with specs, generate images and run published Weave workflows from outside AI tools, and it claims performance changes cut errors and token use. It announced 20 new MCP partner integrations, naming OpenCode, Dia, Junie by JetBrains and Meta Muse, plus an India-first partnership with Sarvam that places Sarvam's Indus alongside tools such as Claude Code, Cursor and GitHub Copilot. Figma also describes its on-canvas agent as now generally available; TechCrunch reported that the agent had reached a wider user base during the preceding week.
The company framed the launches around how AI has changed software work. Loredana Crisan, Figma's chief design officer, told TechCrunch that teams now produce roughly 17 times more code than before but only about 1.3 times more releases, arguing that much of that code is exploratory. India is the strategic backdrop: Figma says the country is its second-largest market by daily active users, that more than 30,000 people registered for the event, and that it will extend its Campus Leaders student programme there from January 2027.
Several caveats apply. Specs, code layers and design-system guidelines are betas, and code sources is in closed beta, so behaviour and pricing may change. The usage numbers, the efficiency gains attributed to the MCP update and the productivity statistics are Figma's own and have not been independently measured. Coverage by TechCrunch, Moneycontrol and Entrackr matches the feature list; only TechCrunch adds original reporting, through its interview with Crisan, and none includes hands-on testing.