The modern design review has acquired a new attendee: an AI agent capable of producing screens faster than a team can say “that is not our brand blue.” Figma is responding with a collection of features meant to tell agents what the company actually intended before the machines decorate the product like an unbriefed hotel renovation. The strategy is less “let the robot design everything” than “give the robot a binder, access to the design system and a politely enforceable sense of shame.”
🤚 The Open-Palm Specification
At its first Config event in India, Figma introduced Figma Specs, a Markdown-based editor for defining a team’s design language. As TechCrunch reports, a spec can live as a standalone Figma file or alongside designs on a canvas; it can reference canvas elements and embed FigJam diagrams, code layers and designs. Teams can also pull in design libraries, code libraries and guidelines so both people and agents have a more centralized reference for the rules.
That distinction matters. A generated interface might look convincing in a screenshot while quietly violating component rules, naming conventions or established interaction patterns. A written instruction such as “make it premium” is not a design system; it is an invitation to gold gradients and existential negotiation. Specs offer a way to put actual constraints closer to the work. They are a reference for agents, not a magical guarantee that an unpredictable model will obey every detail.
Figma has recently made its on-canvas agent available to a wider user base, according to TechCrunch. The new emphasis on guidance acknowledges a familiar problem: AI can generate abundant output, but abundance is not the same as a coherent product. A thousand plausible buttons do not become a strategy merely because they arrive before the meeting ends.
👐 The Two-Handed Prototype
The company is also opening code layers in public beta after announcing a private beta in June. The feature brings code into designs on the canvas for more interactive prototypes. Users can ask an agent to turn a design into code or change a portion of that code while iterating. This is a bridge between a static mock-up and something a team can actually test, not a declaration that generated prototypes are automatically production-ready software.
Figma chief design officer Loredana Crisan framed the imbalance to TechCrunch with a striking comparison: she said there are 17 times more lines of code being produced today than before, but only 1.3 times more releases. Those figures are her characterization, not an independently verified industry census. Her point is that much of the extra code is used to discover what to build, rather than immediately ship it. The idea is plausible enough to make every product manager glance nervously at the prototype folder labeled “final_final_v7.”
Figma also says its MCP integrations now support more than 20 new tools, including Open Code, Dia, Grok Bot and Sarvam, according to TechCrunch. Broader tool compatibility can make it easier to bring agent-driven work into the design environment. It also means teams should be specific about which tools receive context and how human review works. “Our agent can see the library” is a capability statement, not a governance policy.
🌿 The Gentle Awakening
Consider a real use case: a retail team has an established checkout flow, approved components and a marketing campaign landing page due next week. It could put the approved typography, button variants, spacing rules and interaction examples into its shared Figma context, then ask an agent for an interactive checkout prototype. The agent’s first draft is a faster conversation starter; a designer still checks accessibility, content, edge cases and whether the generated flow actually follows the rules. The spec reduces ambiguity. It does not replace judgment, testing or a person willing to say the enormous animated discount wheel must go.
There is a content dimension, too. Figma acquired the open-source headless CMS Payload last year and says it is integrating Payload with Figma Make to generate dynamic websites and manage content directly. TechCrunch reports Figma says Payload downloads have grown sixfold in the past 12 months. That is a company-reported growth figure, not evidence that every generated site is maintainable. The attractive promise is a shorter journey from design and content model to working experience; the less attractive reality is that someone still owns the content, permissions and deployment.
👑 The Gold-Leaf Reckoning
Figma’s bet is that the design tool becomes the room where humans establish intent and agents work inside it, rather than a canvas where successive bots compete to invent entirely new shades of compliance. Specs, code layers and content integration each address a different gap between a plausible picture and a usable product. Their value will depend on how reliably teams can turn guidance into consistent outputs and how carefully they review the results.
There is nothing unglamorous about writing down the rules before inviting machines to improvise. The glamorous option is letting the machines improvise anyway, then paying people to locate the brand afterward. As ever, the premium experience begins with a brief someone actually reads.
“We empowered the agent to innovate, then discovered it had rebranded the checkout as a meditation retreat.” — The Slap of Wisdom Design Governance Salon, quietly restoring the original button