Since 6 October 2026, the Figma agent in Figma Design and the Weave tools are out of beta, and every request now draws on your AI credits. During the beta, none of that usage was counted. If your team got used to asking the agent for five layout variations before lunch, the free period is over, and the way you use it should change with it.
This is not a reason to stop. It is a reason to decide, deliberately, which parts of your design process are worth spending credits on. Here is how the credits work, and a simple way to think about where the agent earns its keep.
What changed on 6 October
Figma's help centre puts it plainly: "We also moved the agent in Figma Design and Weave tools from open beta to general availability on October 6, 2026," and "Usage during the beta period didn't consume any AI credits," according to the Figma help article on the change.
Two details matter for planning:
Weave tools have a set credit cost per tool, so they are predictable.
The agent does not. Its cost depends on how complex your prompt is, what actions it decides to take and how much context you give it. Figma says it cannot predict a request's credit cost in advance.
Both draw from the same pool as Figma Make and other AI features. A pricing tracker, PricingSaaS, logged the change on Figma's pricing page on 12 October, and the launch of Figma Specs in beta on 15 October.
How Figma AI credits work, in plain terms
According to Figma's AI credits help page (checked 19 October 2026), each seat gets a monthly allowance that resets each usage period and does not roll over:
Full seat: 3,000 credits a month on Professional, 3,500 on Organization, 4,250 on Enterprise.
Dev, Collab and View seats: 500 a month on the plans the page lists.
Starter plan: 500 a month, with a daily cap of 150. View seats also have the 150 daily cap.
When a seat's credits run out, Figma uses any shared subscription credits the team has bought, then pay-as-you-go credits if an admin has switched that on with a spending limit. If none of those exist, paid AI actions stop working until the reset or until an admin adds more.
One gap to know about: Figma says granular reporting by feature is not available yet. Admins can see who is using credits and who has hit their limit, but not whether those credits went to the agent, Make or something else.
Where the agent is worth the credits
Picture a typical product design week. Most of the time goes into a few kinds of work. Some of it benefits from an assistant; some of it gets worse.
Good uses: wide, cheap exploration early on
At the very start of a feature, you want breadth. Ask for three different layouts for a settings page, take a quick look, and throw away two. The agent's output does not need to be right; it needs to be different enough to show you options you had not considered. A short, clear prompt with a reference frame usually does this better than a long, vague one.
Good uses: tedious, repeatable edits
Renaming layers, populating a table with realistic content, producing empty, loading and error states for a component you have already designed. These are well-specified tasks where you can judge the result in seconds.
Poor uses: the final 20%
Fine adjustments are where the cost becomes unpredictable. "Make the hierarchy clearer" is a judgement call. The agent may take many actions to interpret it, and you will often redo it by hand anyway. Spacing, type scale and visual weight at the polish stage are faster and cheaper to do yourself, and they are where your judgement as a designer shows.
Poor uses: anything your design system already solves
If you have well-built components and tokens, assembling a screen from them is quick and keeps everything consistent. Asking the agent to generate the same screen risks detached layers, near-miss colours and spacing that breaks your system. You pay for the generation and then pay again, in time, to clean it up.
A simple credit budget for a small team
Because the agent's cost per request is unpredictable, set habits rather than trying to calculate exact numbers.
Check the dashboard after the first week. Admins can see usage and who is at their seat limit on the AI credits dashboard. One week of real data beats any estimate.
Write prompts that point at something. Select the frame, name the component, state the constraint ("use our existing card component, two columns on desktop"). Clear context is the cheapest way to get a useful answer on the first try.
Batch exploration into one session. Ask for variations once, at the start of a feature, instead of nudging the agent repeatedly throughout the week.
Decide on pay-as-you-go deliberately. If you enable it, set a spending limit you are comfortable explaining. If you do not, accept that work may pause near month end.
Give Dev and Collab seat holders realistic expectations. With 500 credits a month, they can try things, but they are not set up for heavy generation.
The design question underneath
The free beta made it easy to treat the agent as a slot machine: pull the lever until something looks good. Metered credits push teams back to a better habit, which is knowing what you are asking for before you ask.
That has always been the skill that separates a strong designer from a fast one. A clear brief, a defined component, a specific constraint. The agent is most useful in the hands of someone who could have done the work themselves and knows exactly which part to hand off. Spend your credits there, and keep the judgement calls for yourself.
