Figma has announced Specs, a Markdown-based editor designed to define a company’s design language and provide a shared reference for people and AI agents. A Specs file can stand alone or sit alongside work on the Figma canvas, according to TechCrunch’s report on the announcement.
The new feature addresses a specific problem created by agent-assisted design: generating an interface is not the same as following a company’s established components, code libraries and guidelines. Figma describes Specs as a way to supply that context centrally. The company is therefore defining how agents should work inside its platform, rather than presenting the agent itself as the sole source of design decisions.
Specs turns design context into an agent reference
A Specs document can reference elements already present on a Figma canvas and incorporate FigJam diagrams, designs and code layers. Teams can also bring design libraries, code libraries and guidelines into the same contextual system, which agents can consult when applying design rules, as described in the TechCrunch coverage.
For a smaller product team, the practical change is the location of the instructions. Instead of relying only on a prompt written for one task, the agent receives a reference connected to the actual design environment. That makes Specs relevant to recurring work such as adapting product screens, preparing campaign pages or keeping interface components aligned across prototypes.
This does not mean the document enforces every rule automatically. Figma told TechCrunch that it is providing steering tools because of the non-deterministic nature of AI and wants to move agent output toward more predictable results. That wording matters: Specs supplies context and direction, while the announcement stops short of claiming deterministic compliance.
Illustration of Specs acting as a shared contextual reference for design teams and agents.
Code layers connect agent output to working prototypes
Figma is also releasing code layers in public beta after an earlier private beta. The feature adds code to designs on the canvas to create more interactive prototypes. Users can ask the agent to turn a design into code or modify part of the code during an iteration, according to the same announcement.
Specs and code layers therefore address different parts of the workflow. Specs holds the intended design language and supporting context. Code layers add executable behaviour to a prototype. Their connection gives product, design and development teams a common place to express both the rule and an implementation that can be explored before release.
For commercial operations teams involved in product launches, the relevant benefit is not autonomous publishing. It is a tighter working artefact for discussions about a page, interaction or product experience. A prototype containing design context and interactive behaviour gives stakeholders something concrete to review, but the source provides no evidence that it removes development work or approval steps.
Illustration of code layers adding interactive behaviour to a design prototype.
The announcement defines direction, not performance
The announcement establishes the availability and intended function of Specs, the public beta status of code layers and the ability to connect design context with agent activity. It does not report comparative tests showing that agents follow brand rules more accurately when using Specs. It also provides no measured results for iteration speed, defect reduction or production readiness in the TechCrunch account.
The distinction is important for teams evaluating the feature. A central reference reduces the need to rediscover scattered guidance, but centralisation alone does not demonstrate that an agent interpreted each instruction correctly. The current evidence supports treating Specs as a structured source of design intent, not as a guarantee of compliant output.
The report also does not explain how conflicting instructions are resolved, how changes to a Specs file affect existing agent work or what review information is exposed after generation. Those unanswered questions determine whether the feature supports a repeatable operational workflow rather than a better-informed individual session.
Payload integration extends the workflow to content
Figma is integrating Payload, the headless content management system it previously acquired, with Figma Make. TechCrunch reports that the integration is intended to support generated dynamic websites and direct content management. Figma has also expanded the external tools supported through its Model Context Protocol connection, according to the announcement.
This broadens the mechanism beyond visual consistency. Specs supplies design context, code layers make prototypes interactive, and the Payload integration connects generated sites with managed content. These are related capabilities, but the source does not demonstrate a complete production pipeline or report an end-to-end customer result.
The next useful evidence will be how Figma documents rule conflicts, agent adherence and the transition from an interactive prototype to released software. Until then, teams should evaluate Specs for one precise decision: whether a shared, canvas-connected reference improves the consistency of agent-assisted design work without treating the announcement as proof of production performance.
