Figma Design Agent Ends Public Beta, Officially Available, Claims New Version Better Follows Commands and Handles Long Tasks
Figma design agent ends public beta and is officially available. The company claims the new version better follows commands and can handle longer tasks; in evaluations scored by professional designers, it outperformed the old version in over 60% of cases.
Library managers can upload markdown guides to the design library, writing rules, best practices, and anti-patterns, which the agent will read each time it is activated in that library. It can also retrieve and insert context from across the platform in conversations, including FigJam notes, Figma Slides copy, and Design boards, and can paste nodes directly into the prompt box. Users can see colleagues' agent actions in real-time on the canvas and can follow by clicking on chat stacks.
Also launched in this batch are: improved latency, generative plugins officially available, the ability to select image models for image editing at the canvas entry, and using the agent to conceive, create, and publish custom skills. Custom skills have been delegated to Starter. Example skills provided by the company include fixing library usage, compliance review, and enhancing component documentation.
Seat rules are separated. Full seats for Professional, Organization, and Enterprise, as well as some Education plans, are available; Starter has limited availability. Collab, Dev, and View seats can only be used in drafts. If an administrator disables AI features for a team or organization, the agent will not appear. The public beta does not consume AI credits, but starting October 6, it will switch to consuming AI credits based on task complexity, with single-point costs not disclosed.
Credits are allocated by seat. Starter has 150 points per day, with a maximum of 500 points per month; Professional's Full seat has 3000 points per month, costing about $16 annually; Organization's Full seat has 3500 points per month, costing about $55 annually; Enterprise's Full seat has 4250 points per month. Figma Make is part of the same credit pool, with a simple text change costing about 30 points, and more complex interactions costing higher; after the agent goes live, it will compete for the same credits as Make.
Buyers are already paying design teams, not new independent subscriptions. The event-driven change is the end of the free public beta and the start of formal billing, with usage shifting from "no point deduction" to AI credits. Beneficiaries are library managers and large teams looking to lock design systems into workflows, while those under pressure are heavy users of Make, Full seats generated by plugins, and independent design generation tools outside the canvas.
Uber, Granola, and Atlassian are listed by Figma as use cases, with scenarios including generating images by library, batch modifying components, and implementing feedback into revisions. The agent is separate from Figma Make: the former modifies design files, while the latter creates interactive applications; external tools can connect to context via verified MCP connectors, while custom connectors require a paid plan.
Source: Public information
ABAB AI Insight
Figma's pricing history starts with occupying the canvas, then shifting to seat pricing. In 2022, Adobe announced a $20 billion acquisition, which was terminated in 2023 due to regulatory issues in Europe and the U.S.; the company subsequently made Dev Mode and organization-level design libraries paid tiers, with AI credits becoming a new unit of measurement after going public in 2025. The design agent was still in public beta in June 2026, and was free for Professional, Organization, and Enterprise Full seats without point deductions; this transition to a formal version effectively recoups the customer acquisition subsidies from that time.
Capital has not flowed into new model laboratories but rather into the existing asset of design libraries. Markdown guides, component variables, FigJam, and Slides are mandated as contexts that the agent must read each time, with library managers uploading once, constraining prompts for the entire team. The motivation is to keep replacement costs within the files: the more teams write specifications into the library, the more the agent can only be used within Figma, rather than exporting drafts to external generation tools. MCP connectors pull in external data like Amplitude, but writing back still requires write permissions; data movement is a two-way gate, not an open exit.
The analogy is not another drawing software but rather Atlassian turning Confluence specifications into a retrieval layer for Rovo, and Adobe embedding Firefly credits into Creative Cloud seats. Figma is currently in a control phase: exploration and batch revisions can be delegated to the agent, but billing, seats, and administrator switches remain under their control. Coding agents like Cursor and Claude can read designs via Figma's MCP but cannot alter the rules file of this library.
This represents a transfer of pricing power. The marginal price of the agent during the public beta was zero, while the official version's price equals task complexity multiplied by remaining credits; Make has long consumed over 30 points per instance, and with the agent's addition, the same 3000 to 4250 points must sustain both revisions and prototyping. The mechanism is that design systems have already centralized, with free trials fostering habits, and the transition to a formal version merely opens the usage scale. Independent generation tools can still compete for drafts, but once team libraries, multi-user following, and credit pools are tied together, the default workflow's pricing power returns to the seats.
ABAB News · Cognitive Laws
Public beta is free to enable formal billing.
Specifications written into documents solidify replacement costs.
Seats sell access, while credits sell usage.