Marcus Andrey Study
Ler em português Library

AI for UX · Step 1 · 9 min read

Figma AI

A beginner's guide to Figma's AI in late 2026: what the agent and the other AI features do, how seats and AI credits work, the data settings to check, and how to try it safely on real work.

Figma's AI is no longer one feature: it's a layer across the whole product. There is an agent you can talk to in a file, small one-click tools, a prototype builder, code on the canvas, and a way for outside AI tools to read and write your files. This note explains each piece in plain words, what it costs, and what to check before you use it on real work.

What you'll be able to do after reading:

  • Explain what the Figma agent, Figma Make, code layers, Weave and the MCP server each do
  • Know which seat and plan you need, and how AI credits are spent
  • Find and understand the data settings that matter
  • Try the agent on a real task and review the result properly

Status as of November 2026. Figma's AI changes every few months: check Get started with Figma AI before you rely on a detail.


1. The big picture

Piece In one sentence Status (Nov 2026)
Figma agent A chat inside your file that creates and edits designs using your design system Generally available in Design since 6 Oct 2026; FigJam and Slides in closed beta
Single-task AI tools One-click helpers: rewrite or translate text, rename layers, search, edit images, add interactions Available
Figma Make Turns a prompt, a design or a code repository into a working prototype or web app Generally available since 24 Jul 2025
Code layers Real, working code placed on the canvas next to your designs Open beta
Figma Weave and Weave tools Generative image and video tools (from the Weavy acquisition, 2025) Weave tools generally available since 6 Oct 2026
Figma Motion Animation in Figma, exportable as video, GIF or animated SVG; the agent can create keyframes Open beta
Figma Specs "A container for your context": the intent and details of a feature, given to agents as structured text Open beta
Figma MCP server Lets outside AI tools (Claude Code, Codex, Cursor…) read your files and write to the canvas Free during the beta; usage-based pricing announced for later
Code Connect Links each Figma component to the real code component, so generated code uses the right one Available on Professional and above
In the fileWork on the canvas

The agent, one-click tools, code layers, Motion.

Where designers spend most of their time
Make thingsBuild and generate

Figma Make for prototypes; Weave for images and video.

Paid in AI credits
ConnectOpen to other tools

The MCP server, Code Connect and Specs link Figma to agents and code.

The back door of the studio
Figma's AI in three groups, as of Nov 2026. Diagram: Marcus Andrey, CC BY 4.0.

2. The Figma agent

The agent is a chat panel in a Figma Design file. You describe what you want, and it works on the canvas:

  • Create: "Make a settings screen for notification preferences, using our components."
  • Edit: "Turn these three cards into a responsive grid and fix the spacing."
  • Explore: "Give me three variations of this empty state."
  • Critique: "What accessibility problems do you see on this screen?"

Since Config 2026 (June) you can give it more context:

  • Skills: saved instructions for tasks you repeat (for example "review against our spacing rules").
  • Connectors: links to other apps, such as GitHub, Atlassian, Slack, Linear and Notion, so it can read a ticket or a doc.
  • Attachments: images, PDFs and other files to work from.

What it can't do yet (Nov 2026): vector drawing, icons, slots and prototyping are listed as "coming soon"; diagrams and charts are sent to FigJam (help article).


3. The other pieces, one by one

Single-task AI tools

Small, fast and low-risk: rewrite or shorten text, translate (useful for multilingual products), rename layers, search with natural language, remove image backgrounds, vectorise, add prototype interactions. Some cost credits; renaming layers and AI search are free.

Figma Make

Describe an app, or start from a design or a code repository, and Make builds a working prototype you can click through and publish. Since Config India 2026, Make has a built-in back end (Payload), so prototypes can store data. Use it for usability tests of complex interactions and to explore ideas early. A Make prototype is not production code.

Code layers

Working code sits on the canvas as a layer: a real date picker, a chart, an animation. You can turn a code layer into editable Figma layers ("extract designs") and push changes back. Useful when a static mock-up can't show a state or an interaction.

Figma Weave and Weave tools

Image and video generation inside Figma: about 37 first-party tools, each with a fixed credit cost. Note that Figma's content-training policy does not apply to Weave; it has its own terms, so check them separately.

Figma Specs

A place for the "why" and the details of a feature, written so that agents can read it. It connects to the idea in Design specs for AI: when AI can draw screens, the spec becomes the most important thing you write.

The Figma MCP server and Code Connect

MCP is a standard way for AI tools to connect to other apps. Figma's MCP server lets a coding agent read your design (layers, variables, screenshots) and write to the canvas. Code Connect tells it which code component matches each Figma component, so the code it writes uses your real components instead of look-alikes. See Generating design with Claude Code.


4. Seats, plans and AI credits

Seats decide what you can do:

  • In Figma Design, you need a Full seat to use AI in shared files. Any seat can try AI in your own Drafts.
  • With a Dev, Collab or View seat, you can ask the agent questions but not edit.
  • Writing to the canvas through the MCP server needs a Full seat, outside Drafts.

Plans decide which features exist: most standalone AI tools (Make, Weave tools) need Professional, Organization or Enterprise.

AI credits are a monthly allowance (Figma help):

  1. Full · Starter500
  2. Full · Professional3,000
  3. Full · Organization3,500
  4. Full · Enterprise4,250
  5. Dev, Collab, View500
Figma AI credits per seat per month, from Figma's help centre (Nov 2026). Credits don't roll over; Starter also has a cap of 150 a day. Diagram: Marcus Andrey, CC BY 4.0.

Credits don't roll over to the next month. Image generation costs vary by model, from 2 to 16 credits per image in Make, and Weave tools cost 13–110 credits per run. A long agent session can use a lot: watch your balance before a deadline.


5. Data settings to check

Setting Default Who can change it
Content training (Figma may use your files to improve its AI) On for Starter and Professional; off for Organization and Enterprise Admins on Starter and Professional can turn it off
AI on or off for the whole organisation On Only Enterprise admins can switch AI off (with per-workspace exceptions)
Agent chat visibility Visible to Full-seat editors of the file Each chat can be made private
Weave and Weave tools Separate terms Check Weave's own policy

Source: Manage AI settings and content training.


6. Using the agent well

  1. Start from your design system. Ask the agent to use your library's components and variables. Output that ignores the system costs more to fix than it saves.
  2. Give a brief, not a wish. Who is the user, what is the task, what states are needed (empty, loading, error), what constraints apply. See Prompt patterns for designers.
  3. Work in small steps. One screen or one change at a time, then look.
  4. Review like a colleague's work. Check components, spacing, contrast, copy and states. See AI design review process.
  5. Keep the decisions. The agent proposes; you decide, and you can explain why.

Common mistakes:

  • Accepting the first result because it looks finished.
  • Detached or look-alike components that break the design system.
  • Placeholder copy that ships ("Lorem ipsum", invented numbers).
  • Pasting confidential information into a shared chat.

7. Try it (no code needed)

  1. Open a copy of a real file in your Drafts (so nothing shared changes).
  2. Ask the agent for three variations of one screen, using your components.
  3. Review each one with five questions: right components? right spacing? readable contrast? sensible copy? all states?
  4. Write down how long it took, including the fixes, and whether you'd use the result.

Reflection questions: Which part of your work did the agent speed up, and which part did it slow down? What would you need to give it (spec, examples, rules) to do better next time?


8. If you lead a team

  • Ask your Figma admin: which plan do we have, is AI on, is content training off, are Weave tools allowed, how many credits per seat?
  • Set up Code Connect for the core library: it's what makes AI-generated code use your real components.
  • Decide where specs live (Figma Specs, a wiki, or the code repository) and say so in the team guidelines.
  • Track credit use for a month before deciding who needs which seat.

Checklist

Learn more

Books, courses and videos to go deeper

AI tools landscape for designers · Claude Design · Generating design with Claude Code · Design specs for AI · AI design review process · MCP · Design tokens

Saved only on this device.

↑ ↓ to move · Enter to open · Esc to close