Figma Specs teaches AI agents a brand's design language

At its first Config India event in Bengaluru on 15 October, Figma introduced Specs, a Markdown editor where teams write down their design language for both human designers and AI agents. Code Layers moved from private to public beta. India is Figma's largest market by users outside the US.

Specs can be standalone files or sit on the canvas, linked to elements, FigJam diagrams, code layers and designs. Teams combine design libraries, code libraries and written guidelines into one context that agents consult, so an agent building a checkout flow uses the brand's real buttons. Figma's MCP support added more than 20 tools, and its Payload CMS now connects to Figma Make.

Markdown is plain text formatting that people and models read easily. A design system collects a brand's colours, fonts, components and rules. MCP lets AI models connect to external tools in a standard way.

Agents generate interfaces quickly, but often off-brand, with slightly wrong colours and reinvented components. Specs puts the rules in one machine-readable place, mirroring the Markdown rule files now standard for coding agents.

Code layers bring real, working code components onto the design canvas, shortening the hand-off between design and development. Connecting Payload, the headless CMS Figma acquired, to Figma Make eases the path from design to a content-managed site. Figma says Payload downloads grew sixfold in 12 months.

Designers' work is shifting from drawing screens to writing the rules agents follow. Brands with well-defined systems will get consistent output; others will fix every result by hand.

Rival tools are expected to support similar rule files, so writing design rules in plain text will pay off whichever tool you choose. Specs pricing and general availability have not been announced.

Write a one-page brand guide in plain text covering colours, fonts, logo use and tone, name your recurring components, and check every AI-generated piece against the same checklist.