Figma Wants to Own the Instructions, Not Just the Canvas

Figma’s biggest threat isn’t a better design tool. It’s a coding agent that makes opening Figma optional.

That’s the useful lens for its October Config India announcements: Specs, code layers in open beta, design-system guidelines, and a native Payload backend. These aren’t disconnected AI features. They’re Figma’s bid to remain the place where teams decide what software should do—even when an agent writes the interface.

Smart strategy. But let’s separate shared context from actual control.

The Real Story

Generating a convincing screen is getting cheap. Preserving the decisions behind that screen is still expensive.

Figma Specs addresses that gap with documents containing requirements, decisions, code blocks, images, and embeds. Specs lives independently or inside Figma Design and FigJam, and is entering open beta through a phased rollout. Through Figma’s Model Context Protocol server, agents using the figma-spec skill can read, create, and update those documents.

That changes the handoff. An agent can work against persistent product intent instead of whatever somebody remembered to paste into a prompt.

CEO Dylan Field’s strategy of bringing code, agents, and creative work onto one canvas fits this direction. Figma wants external agents participating in its workspace, not replacing it.

Your Design System Is Becoming Agent Input

Library guidelines supply agents with conventions and anti-patterns when they use the associated library. That fills a real hole: a component tells an agent what exists, but rarely explains when using it is a terrible idea.

Figma recommends three to eight focused guideline files totaling under 100 KB, with a hard combined limit of 200 KB per library. Supported formats include Markdown, plain text, and JSON.

Good. Nobody needs a 400-page brand manifesto in every generation request.

But guidance is not enforcement. “Never use raw color values” is an instruction. A lint rule rejecting raw color values is a control. Keep both.

Availability also needs a sober reading:

• Guidelines: open beta.
• Code sources in libraries: closed beta, connecting repositories or npm packages.
• Centralized design systems: preview, not a finished governance platform.

Treat instructions like engineering assets: assign owners, review changes, and test representative tasks. Otherwise, you’ve automated the distribution of stale documentation. Congratulations, it’s faster now.

Running Code Beats Another Click-Through Demo

Code layers in Figma Design are moving into open beta. They contain running code, with starting points including design frames, prompts, GitHub repositories, uploaded folders, and captured webpages.

The interesting part is the loop: extract a running interface’s state or recorded flow into editable design layers, refine it, and send changes back. Figma also describes pushing finalized changes to a repository.

That’s useful for reviewing behavior rather than admiring screenshots.

It isn’t frictionless. Imports can take 10 minutes or more, some projects need modifications, and creating imported code layers requires a Full seat. A canvas preview doesn’t establish maintainability, accessibility, or secure authorization.

Payload adds schema-driven data, authentication, access controls, storage, and APIs as the native backend for Make and code layers. Existing Supabase-backed Make apps continue working. This is application-building territory, not merely nicer prototyping.

The Meter Runs Behind the Canvas

Code-layer generation and editing are free on paid plans through November 4, 2026; other agent work in the same prompt can still consume credits. Afterward, code-layer tasks consume AI credits based on complexity.

There’s a business model here. Figma reported second-quarter 2026 revenue of $370.1 million, up 48% year over year. CFO Praveer Melwani called it the first full quarter of AI-credit monetization. Those results predate this release.

I’d pilot this with a real repository and measure review effort, guideline adherence, and cost per accepted change. Generated code volume is a vanity metric.

Figma is making a credible bid to own the instructions. Engineering teams still own whether the resulting software deserves to ship.