On October 6, Figma moved the agent in Figma Design out of open beta and into general availability. The same day, the agent and the Weave image tools started spending AI credits. The release notes list faster responses, file search in chat, and live views of teammates’ agents. They also list the feature I think matters most for anyone who owns a brand: guidelines.
Guidelines are plain text files you attach to a published design library. Figma’s help center says that when someone asks the agent to design with that library, the files are “automatically provided to the agent in the background.” Your brand rules stop being a PDF people skim once and become context a machine reads on every prompt. That is useful, and it has a price that grows with how much you write.
A note on evidence before going further. This is a primary-source review. I have not run the agent on a paid Figma account for this piece. Everything below comes from Figma’s release notes, two Figma blog posts (May 20 and October 6), the help center articles on the agent, guidelines, Weave tools, and AI credits, and Figma’s pricing page, cross-checked against The Verge, Creators Toolbox, and Istar’s credit guide. Where Figma’s own pages disagree, I say so.
What it is
The agent lives inside Figma Design, in the desktop app and the browser. You select a layer and open a prompt box on the canvas, or you open the Agents panel in the left rail and chat in a sidebar. From there it can generate screens, restructure layouts, apply your components, styles, and variables, fill frames with realistic copy and images, rewrite or translate text in bulk, edit images, sort comments, and critique a design “from the perspective of different personas.” It runs prompts in parallel and connects to outside tools over MCP.
Figma announced it on May 20 as a gradual closed beta for Full seats on Professional, Organization, and Enterprise plans, with Collab and Dev seats limited to drafts and Starter, Education, and Government plans left out. Beta use was free, with credits promised at general availability.
For marketing teams, the useful part is the production grind: swapping copy across forty frames, localizing a campaign set, populating templates, and running Weave tools on images. Weave tools are packaged image apps, such as Generate mockup (Figma’s help center suggests “an out-of-home campaign or product packaging”), Replace background, Change lighting, Transfer style, and Product shoot.
What changed on October 6
Three things moved at once.
The meter started. Figma’s credit FAQ says the agent and Weave tools “now consume Figma AI credits” and draw from the same budget as Figma Make. Usage during beta cost nothing. Weave tools have a fixed price per run. The agent’s cost varies with the prompt, the actions it takes, and the context you give it.
Guidelines arrived. You can now upload Markdown, plain text, or JSON files to a published library with rules, best practices, and anti-patterns. The agent reads them every time someone prompts with that library enabled. The Verge and Creators Toolbox both highlighted it on October 6.
Access widened and got more social. Figma’s October 6 blog post says the agent is available to Full seats on Professional, Organization, Enterprise, and some Education plans, with “limited availability on the Starter plan,” and that Collab, Dev, and View seats can use it in their drafts. Custom skills, which are reusable packaged workflows, are now available on Starter. Chats started in a file are visible by default to anyone with a Full seat and edit access (that default dates to June 23), and you can now follow a teammate’s agent live on the canvas.
What works
Figma’s guidance on writing guidelines is genuinely good. The help article tells you to focus on where people get it wrong, write scenario-based dos and don’ts with the reasoning behind them, and give a clear recommendation rather than a menu of options. It suggests starting with color, spacing, and corner radius because those change output the most. It even supplies a drafting prompt that tells the agent not to invent component names, tokens, or rules and to mark missing information as a question. That is how a brand book should have been written all along.
The bulk work is the right target. Figma’s October 6 design systems post describes an Uber designer using custom skills to generate component specs across seven platforms, and a Granola designer asking the agent to swap an imported button for the matching system component. It also shows the agent auditing a screen and flagging a hand-built button and a hard-coded color. For marketing design, the equivalent is checking a landing page or a social template set against the brand library before it ships.
Weave tools have a known price. Unlike the agent, each first-party Weave tool lists a fixed cost per run as of October 6: 19 credits for Generate mockup, Replace background, and Add logo to product, 21 for Product shoot, 78 for Sticker pack, and 110 for Generate product variations. Those can be budgeted like stock photos.
The admin plumbing exists. Seat credits are spent first, then a shared subscription pool if you buy one, then pay-as-you-go up to a monthly cap. Admins can set custom limits per person or turn off paid credit access entirely. Enterprise plans get an AI Usage API with per-user, per-day data for chargeback.
Where it breaks or is fenced
Your rulebook costs credits on every prompt. The guidelines article says plainly that “a prompt may use more AI credits when it uses a library with guidelines,” and that repeated information “can increase processing time and AI credit use.” Pasting a sixty-page brand book into a library is the expensive way to do this. Figma recommends three to eight focused files under 100 KB combined, with a hard ceiling of 200 KB.
You see the price only after you pay it. Figma says it cannot predict what an agent prompt will cost in advance. You hover over the credits icon after the task finishes. Its own approximate examples are 30 or more credits to change the fonts in a design, 100 or more to summarize a canvas, and 300 or more to generate a five-screen mobile checkout flow. Note the plus signs. Undoing an agent change reverts the file but does not refund the credits.
You cannot see which feature spent the money. The admin help page says “granular reporting by feature is not available at this time.” The agent, Figma Make, and Weave tools draw from one pool, and the dashboard reports credits per user, not per feature. The one per-product breakdown that existed, a beta usage CSV for Organization and Enterprise admins, stopped collecting new data on October 6.
Paid credits are open to everyone by default. “All users have access to additional purchased AI credits by default,” per the same page. If you buy a subscription first and set limits second, every seat on the plan has a tab open in the meantime.
Annual plans lock the credit purchase. On Organization and Enterprise, the AI credits subscription is annual and can be downgraded or cancelled only in the 30-day window before renewal. Professional is month to month.
Guidelines are fenced. They require a Professional, Organization, or Enterprise plan, edit access, and a published library. They are not used when creating a Make kit, and Figma Make keeps its own guidelines inside each Make file, so a team using both now maintains two rulebooks. The help article also still says that “during beta” you write and upload the files yourself.
It does not finish the job. The help center’s feature table lists exporting assets as unsupported, with vector editing, icon creation, and prototyping marked “coming soon.” The agent works only in Figma Design for now; FigJam and Slides versions are in closed beta. Weave tools are unavailable to customers with the Governance+ add-on. The agent can make the forty resized variants, but a person still exports and ships them.
Design feedback is a simulation. The agent can critique a design as a VP of Product would, or “suggest opportunities to improve conversion.” That is a decent pre-crit, not a test result.
Visible chats change what you type. New chats in a file are visible by default to Full seat editors. On a shared client file, that includes whatever background an agency pasted into the prompt.
Figma’s pages disagree on Starter. The May launch post excluded Starter. The October 6 blog says “limited availability.” The help article on using the agent is labeled “available on all plans.” Check the actual account before you plan a freelancer’s workflow around it.
Who it is for right now
It fits best for in-house brand and design operations teams on Organization or Enterprise plans who already have a published library and a person who owns it. The jobs that pay off are high-volume adaptation: localizing a campaign set, populating templates, swapping copy and imagery at scale, auditing marketing pages against the system, and making product mockups with Weave tools at a known cost.
Agencies that run a library per client fit too, because guidelines attach per library. Settle chat visibility first.
It fits less well for Starter users (500 credits a month, 150 a day, no guidelines), for teams without a published library (Figma’s May post says the agent by default starts from your most frequently and recently used components), and for anyone expecting finished, exported campaign files.
What it costs
There is no separate price for the agent. It spends the seat credits in your plan. Figma’s pricing page lists a Professional Full seat at $16 a month with 3,000 credits, an Organization Full seat at $55 a month billed annually with 3,500, and an Enterprise Full seat at $90 a month billed annually with 4,250. Dev, Collab, and View seats get 500 a month. Starter is free, with 150 credits a day up to 500 a month. Seat credits reset monthly, do not roll over, and cannot be shared.
Beyond that, admins can buy a monthly credit package, where the price per credit falls as volume rises, or turn on pay-as-you-go with a monthly spending limit. Figma does not publish those rates on its public pages; they appear in the admin dashboard. On August 25, Figma raised the credits in existing add-on packages at the same price, 2x on Professional and 1.6x on Organization and Enterprise, six weeks before metering began.
The rough arithmetic: using Figma’s own estimate of 300 or more credits for a five-screen flow, a Professional Full seat covers at most about ten of those a month before it touches a paid pool. On Starter or a View seat, the 150-credit daily cap is smaller than that single estimate.
What to do this week
Treat October 6 as two changes: a new creative ops task (the rulebook) and a new spend line (the meter).
- Pull the beta CSV today if you are an Organization or Enterprise admin. It is the only per-product view of usage Figma currently offers, and it stopped updating on October 6. Use it to size the first month.
- Set per-person credit limits before you buy anything. Purchased credits are open to all seats by default.
- Write a lean guidelines pack for your main brand library. Three to five files, well under 100 KB: color roles, type and spacing, and an anti-patterns file written as scenarios with what to do, what not to do, and why. Use Figma’s drafting prompt, then edit hard and cut anything repeated.
- Run one real adaptation job as a metered pilot. One campaign’s frame set, localized into two languages. Run it once with the library and guidelines attached and once without, hover the credits icon after every prompt, and log task, context, credits, and how much a designer had to fix.
- Budget Weave tools per campaign. They have fixed prices, so give each campaign a run count instead of an open tab.
- Set a chat rule for client files. Confidential briefs stay out of agent chats in shared files, or the work happens in drafts.
- Hold off on an annual credit subscription. On Organization and Enterprise you cannot downgrade until renewal, so wait for a full month of real usage.
Figma’s agent is good at the thing brand teams never have time for: making the hundredth variant follow the rules. It only knows the rules you write down, it charges more the more you write, and it cannot tell you afterward which feature spent the money. Write the rulebook short, put a limit on the tab, and measure one real job before the next credit reset.
