The short answer
Figma AI is the set of AI features inside Figma, made by Figma, Inc. The main ones are the Figma agent, which edits designs on the canvas and became generally available on 6 October 2026, and Figma Make, which turns prompts and designs into prototypes and web apps. Both use AI credits: a Full seat gets 3,000 a month on Professional, 3,500 on Organization and 4,250 on Enterprise, while the free Starter plan gets 500.
| Made by | Figma, Inc., United States |
|---|---|
| Free plan | Starter, with 500 AI credits a month and a 150 credit daily limit |
| Models | Not all named by Figma; Gemini 4 Argon is in the Figma Make model picker |
| Best at | Design exploration, bulk edits, design system documentation, prototypes |
| Shows sources | Not applicable; the agent can search the web for data |
| Team plans | Professional, Organization and Enterprise, with admin credit controls on Enterprise |
| Your data | Content training is on by default for Starter and Professional, off for Organization and Enterprise |
What Figma AI actually is in late 2026
Figma, Inc. started adding AI to its design tool as a set of small helpers, such as renaming layers and finding components. In 2026 it became something bigger. There are now three main pieces worth knowing about, and each one suits a different job.
The Figma agent works directly on the design canvas. You ask it, in plain language, to explore design directions, make bulk edits, document a design system or apply a round of feedback, and it changes the file in front of you. It launched in beta on 20 May 2026 and became generally available on 6 October 2026, which is also the day it started using AI credits.
Figma Make turns a prompt or an existing Figma design into an interactive prototype or a working web app. It has a model picker, and since 13 October 2026 Gemini 4 Argon has been rolling out there on all plans.
The Figma MCP server connects Figma to outside AI tools, so coding agents can read your designs and act on them. Figma shows OpenAI, GitHub, Claude, Cursor and Gemini among the tools you can connect.
Around those sit Weave tools for generating images and vectors, generative plugins and shaders you can build by prompting the agent, and the everyday helpers: erase objects, remove backgrounds, upscale, vectorise, rewrite text and search visually.
What the Figma agent and Figma Make are best at
Exploring several directions in a real file
The agent's strongest use is breadth. Select a screen, ask for three alternative layouts for a different audience, and compare them side by side using your own components. Because it starts from your file and your design system, the results are closer to usable than a picture from a general image tool.
Boring, repetitive edits
Renaming hundreds of layers, applying a new spacing rule across a set of frames or writing documentation for every component in a library are tasks people put off for weeks. The agent handles them in minutes, and you review the result instead of doing the clicking. Since the June 2026 update you can also save a prompt you repeat as a skill, a reusable slash command, and publish it to your team.
From design to something clickable
Figma Make is the quickest way to show stakeholders how a flow behaves rather than how it looks. Product managers and marketers who do not code can turn a design into a prototype people can actually tap through, which makes usability testing and sign-off much faster.
Bringing context in
The agent can read attached files such as CSVs, follow pasted Figma links, search the web and connect to GitHub, Atlassian, Slack, Linear and Hex. A Google Drive connector for the agent and Figma Make arrived on 30 October 2026 on all plans, including Starter. That means you can ask it to fill a dashboard design with the real numbers from a sheet instead of placeholder text.
Where Figma AI falls short
- Credits are hard to predict. The agent, Make and code layers use a variable number of credits depending on the model, the task and the context you give. Figma's own examples run from about 30 credits for a small edit to more than 300 for a multi-screen flow.
- One shared budget. All AI features draw from the same monthly allowance. If you spent your credits in Make, the agent may show zero left.
- Little AI on the free plan. Starter gets 500 credits a month with a 150 credit daily cap, and the Figma agent is not available on Starter at all. On paid plans it is for Full seats, with Dev and Collab seats limited to drafts.
- It does not replace design judgement. The agent produces plausible options fast. Deciding which one solves the user's problem is still your job, and generated prototypes need checking before anyone treats them as production code.
How to get better results from the Figma agent
- Start from a selection. Select the frame or layer you mean before you prompt. The agent can start from any layer, and a clear starting point beats a long description.
- Name your design system. Tell it which library, components and tokens to use. Since 15 October 2026 Guidelines, in open beta, let you add design system guidance to published libraries for the agent to follow, so you do not repeat yourself.
- Ask for options, not one answer. Request three directions with a sentence on the thinking behind each. Comparing is faster than correcting a single attempt.
- Keep prompts small. Because cost grows with the work the agent does, five focused prompts are usually cheaper and easier to review than one huge request.
- Save what works as a skill. When a prompt gives a good result, turn it into a skill so the whole team gets the same quality.
New to prompting? Our Figma lessons and the guide to writing a good prompt are short and practical.
Figma AI privacy settings and where to find them
Figma separates two things: access to AI features and content training, which decides whether your files may be used to train AI models. On Starter and Professional teams, content training is on by default. On Organization and Enterprise plans it is off, and Figma says it cannot currently be switched on.
To switch it off on a Professional team, open the file browser, open your team, choose Admin in the left sidebar, then the Settings tab, and switch off Content training in the AI section. On Starter, open the menu next to your team name, choose View settings, then Settings, the AI section and Manage AI settings. Turning it off covers new content and edits from that point on, and it does not remove your access to AI features.
Which Figma plan suits whom
Each Full seat comes with a monthly credit allowance: 3,000 on Professional, 3,500 on Organization and 4,250 on Enterprise. Dev and Collab seats get 500. Credits reset monthly and do not roll over, and admins can buy add-on credits.
- Freelancers and small studios: Professional with a Full seat is the entry point for the agent and full Make. Switch content training off on day one if client files are involved.
- Product teams in larger companies: Organization or Enterprise, mainly because content training is off and Enterprise admins can split credits by team and track usage through an API.
- Developers and reviewers: a Dev or Collab seat is enough to view, comment and try the agent in drafts.
- Not for: anyone who only needs a quick image or social graphic. A tool like Canva or Adobe Firefly is simpler for that.
Prices change, so check Figma's pricing page before you buy, and see the best AI for designers for how Figma fits next to other tools.
Prompts to copy
Three directions for one screen
Using the selected frame as the starting point, create three alternative layouts for [audience, for example first-time users on mobile]. Use only components from our [library name] library. Under each one, add a short note on what it prioritises and what it gives up.
You get: Three on-system options you can compare side by side
Document a component library
For every component on this page, write usage documentation: what it is for, when not to use it, the variants available and one accessibility note. Put it in a text frame next to each component and keep each entry under 80 words.
You get: Consistent documentation for the whole library in one pass
Apply review feedback
Here is the feedback from our design review: [paste comments]. Group it into themes, tell me which points conflict, then apply the uncontroversial changes to the selected frames and list what you left for me to decide.
You get: Feedback applied, with the hard calls flagged
Fill a design with real data
Use the attached CSV to replace the placeholder content in this dashboard design. Keep the layout, format numbers as [currency or percentage], and highlight any value that does not fit its container.
You get: A realistic mock-up with layout problems spotted
Turn a flow into a prototype in Figma Make
Build a clickable prototype from these three screens: [screen names]. The user should be able to [main task, for example sign up and set a goal]. Add realistic empty, loading and error states, and keep our colours and type styles.
You get: A prototype stakeholders can tap through, including edge cases
A design QA check to save as a skill
Check the selected frames for inconsistent spacing, colours that are not in our [library name] colour styles and text below 14px. List each issue with the layer name and a suggested fix, grouped by frame. Do not change anything yet.
You get: A QA report you can save as a skill so anyone on the team can run it as a slash command