On 27 June 2026, a Figma Professional subscriber named Zafar Ali posted on Figma’s forum that he had “created only 6 design pages using Figma Make” and that “those 6 pages consumed all 3,000 AI credits”. One later prompt, he wrote, used 1,000 credits at once. A Figma moderator, Tom Reem, answered the next day that Make “uses agentic AI, meaning the model determines what actions” it takes. On 6 October the Figma agent inherits that logic. Figma sells the agent as a way to try many design ideas. From that date it charges for every attempt, can’t say the price beforehand, and doesn’t refund the ideas you throw away.
The meter is built on rules Figma states plainly
Per Figma’s help page on AI credits , “because the AI determines what actions it needs to perform in order to complete the task, we can’t predict exactly how many credits it will use.” The cost appears only afterwards, by hovering over a credits icon under the finished prompt. And “undoing an AI action reverts changes in your file but does not refund credits, since the task has already been processed.”
The AI credit updates FAQ confirms the date. The agent in Figma Design moves from open beta, which “is free and doesn’t consume any AI credits”, to general availability on 6 October, drawing from the same pool as Make. A Professional full seat gets 3,000 credits a month, per Figma’s credits explainer .
For scale, Figma’s approximate Make rates as of February 2026 were about 30 credits for a font change, 75 for adding interactivity and 100 for generating an app from scratch. Those are Make figures, not agent rates.
Exploration is the expensive part
Figma’s launch post for the agent organizes the agent around three jobs, headed “Explore more directions”, “Automate busywork” and “Do more with feedback”, and pitches it as a way to generate several stylistic approaches at once. Exploring means producing several things and keeping one.
Run the arithmetic yourself. Ask the agent for four stylistic directions on a settings screen, keep one, and undo the other three. Under the stated rules, all four were billed, you saw the cost only after the run, and the undo returned nothing. Three-quarters of that session is billed as waste.
A meter that can't quote a price and never refunds a discard charges most for the work the agent is sold for.
Software Pricing Partners makes the general case in a field guide on usage pricing for AI : the model “makes a customer’s exploration visible as a cost line”, and finance then “governs that cost line by cutting the activity.” Its cases include Replit’s meter billing failed agent runs, so, in the guide’s words, “the bill rises exactly where the value fell”.
Here the leap is ours. None of these sources measured designers under the Figma agent’s meter, and nobody can yet, because billing starts on 6 October. Read across from the billing rules and from developer tools, we expect designers to try fewer directions. That is a projection.
What a visible cost might do instead
Think of a product designer who spent the beta trying four layout directions before keeping one. Before each prompt, she now decides whether three discards are worth paying for from an allowance shared with Make. She can’t know the price first. The credit balance becomes a second input beside her judgment, and the sensible response to an unpriced meter is to prompt less.
There is a fair case against this. Nielsen Norman Group’s 2 October study of non-engineers building with AI found that unlimited building invites excess: “Because AI makes it so easy for you to build something, people actually start overbuilding because it feels productive,” one participant said. On that reading, a visible cost could make exploration more deliberate rather than just smaller. Figma has also doubled credits on Professional add-ons at the same price since 25 August, per its FAQ.
The pricing consultancy cuts both ways. It says usage pricing works when use cases are proven and consumption predictable. That describes the agent’s bulk edits, such as renaming and component swaps, far better than open-ended exploration.
A meter that can’t quote a price and never refunds a discard charges most for the work the agent is sold for. It probably suits chores and strains exploration. Friction can sharpen a brief, but it can’t tell you whether the fourth direction would have been the good one. As we noted in our look at an exploration-tuned design agent , breadth is hard to score even when it is free.
The price per credit will move, and Figma may soften the rules. The shape is what to watch: a bill that arrives after the draft, for drafts you were meant to throw away.
