Figma Gives AI Agents a Spec Sheet and Developers a Shared Canvas

AI coding tools have made building software cheap. They have not made it cheap to decide what software to build. A batch of updates Figma rolled out this week at its first Config event in India addresses that gap.

The company moved code layers into open beta, launched a Markdown-based tool called Specs, expanded its Model Context Protocol (MCP) server, and tied its Payload headless content management system into Figma Make. Together, the updates aim to put designers, developers, product managers, and AI agents on the same canvas, working from the same context.

Figma’s chief design officer, Loredana Crisan, put the problem in numbers. She told TechCrunch that teams now produce 17 times as many lines of code as before, but only 1.3 times as many releases. She said much of that code exists to figure out what the product should be rather than to ship.

That will sound familiar to anyone running a delivery pipeline. And it raises a fair question. If most AI-generated code is exploration, where should that exploration happen? And how does the good stuff reach production without losing the reasoning behind it?

Code on the Canvas

Code layers let teams place running, code-backed prototypes directly in a Figma file. Code can come from a GitHub repository, a local folder, a live website captured with Figma’s Chrome extension, or a prototype built with an AI coding tool. Teams can also start with a design frame and ask Figma’s agent to turn it into working code.

It works in both directions. A team can pull an existing coded interface onto the canvas and extract screens or states back into editable design layers. Collaborators can then compare versions, comment, and refine behavior together, instead of reviewing static mockups or screen recordings.

Figma first released code layers in private beta in June. Creating and editing them is free on paid plans through Nov. 4, 2026. After that, AI credit usage may apply.

Specs as Context for Agents

Specs may matter more to DevOps teams. It’s a Markdown editor that lives on the canvas or as a standalone file. Teams use it to write down requirements, decisions, design rules, and instructions for agents. A spec can reference canvas elements and embed FigJam diagrams, designs, and code layers. It can also pull in context from tools like a connected GitHub repository.

The point is predictability. AI agents are non-deterministic. Give one a short prompt, and you get a guess. Give it documented rules, a design library, and a code library, and you improve the odds it builds something that fits. Figma also announced Guidelines, in open beta on Professional, Organization, and Enterprise plans, to help teams define those rules. A feature that links design system components to their source code is in closed beta.

Specs is rolling out in open beta on web, desktop, and mobile for paid plans.

But a spec is only as good as its accuracy. Mitch Ashley, vice president and practice lead for CIO & Technology Buyers and Software Lifecycle Engineering at The Futurum Group, sees that as the core risk.

“Work is moving from tool interfaces to where the developers work, agentic dev surfaces,” Ashley said. “And a spec only helps if it stays true to the code it describes. Agents change code faster than people update documents, so drift between intent and implementation will show up first in review queues and incident postmortems.”

More Agents, More Connections

Figma’s MCP server now lets agents create code layers, read and write Specs, convert agent-generated HTML into editable design layers and generate images. The company says it supports more than 20 additional tools, including Open Code, Dia, Grok Bot and Indian AI startup Sarvam. Figma previously added the ability to send work from Claude Code into Figma and pull design changes back into a codebase through MCP.

Payload, the open-source headless CMS Figma acquired last year, now works inside Figma Make. Teams can build dynamic sites with real content and data instead of placeholder text. That helps surface problems early, like layouts that break with long headlines. Figma says Payload downloads grew sixfold over the past 12 months.

What It Means for DevOps

The design-to-development handoff has always been leaky. A study titled “Design Breakdowns: Designer-Developer Gaps in Representing and Interpreting Interactive Systems” found that common practices create needless rework. Based on 16 interviews, researchers found designers often left out critical details, missed edge cases, or overlooked technical limits.

AI agents don’t fix that on their own. They can make it worse by producing more code, faster, with less shared understanding. Figma’s bet is that context, not code, is now the scarce resource. In that sense, Specs applies the “everything as code” idea to intent.

Ashley advises managing it that way. “DevOps teams should put specs under the same change control as code: versioned in the repo, owned by a named team, and reviewed alongside the code they drive,” he said. “Treat the record of what an agent was told to build as audit evidence.”

There are other limits. A prototype on a canvas is not a production service. It still needs testing, code review, security scanning, and integration work. Large codebases carry dependencies and architecture decisions a design canvas can’t fully show. And once the promotional period ends, AI credits will shape how freely teams experiment.

The Bottom Line

Figma wants the canvas to be where teams decide what to build before pipelines take over. For DevOps teams, the most useful piece may not be code layers. It may be the less exciting one: A shared record of what an agent was told to build and why. Whether that record stays accurate, and who owns it, will decide how much it helps.