Figma just hired a new team member. Title: AI Agent. Start date: May 20, 2026.
Here's the thing — this team member works directly on your canvas, inside your actual design file. Tell it to "explore three different directions for the checkout flow" and it actually does it. All at once.
Wait, doesn't Figma already have AI tools?
It does — First Draft for quick layouts, Rename Layers, background removal, text rewriting. But all of those are button-press tools. You click, it runs, done.
The new Figma design agent is fundamentally different. It lives on your canvas as a persistent collaborator — working inside the same file as your team. Figma's own description says it best: "A dedicated design agent that works beside you on the canvas, inside the same file as your team — as a true collaborator."
It runs on models fine-tuned specifically for design work. It understands layout, components, visual hierarchy, and design tokens. And you can run multiple agents in parallel — one generating concepts while another cleans up component naming conventions.
Context matters here. Figma reported Q1 2026 revenue of $333.4 million — 46% year-over-year growth. At the same time, it's facing real pressure from Canva, Adobe Firefly (41% enterprise adoption), and AI-native newcomers like Flora and Krea. After partnering with Anthropic and OpenAI to bring Claude Code and Codex into Figma, this is Figma's own bet: a native agent built specifically for design.
CDO Loredana Crisan framed it well: "As building software gets easier, what matters most is setting direction — deciding what to work on, how it should function, what the experience should feel like." The agent handles the execution details. You focus on direction.
So what can you actually ask it to do?
Three categories, according to Figma's launch docs.
Explore more directions, simultaneously
Early-stage design work is expensive — all that back-and-forth comparing "direction A vs direction B." The agent handles this in parallel. Prompt it to "explore three checkout flows with different color palettes" and all three appear on the canvas.
The killer feature here: @mention your design system components, color tokens, and variables directly in the prompt. "@Button/Primary using @Color/Brand tokens" — it'll use exactly what your team defined, not some generic approximation.
Automate the repetitive stuff
Swapping a component across 200 screens. Renaming every variable in the file for consistency. Applying the same padding update across 50 frames. These used to mean plugins or just grinding through it. Now it's one sentence.
Apply feedback at scale
Hand the agent a comment thread with 30 pieces of feedback. It'll organize the themes and translate them into design revisions. You can even ask it to stress-test the design from a different stakeholder's perspective.
| Before | Figma Design Agent | |
|---|---|---|
| Exploring concepts | One at a time, sequentially | Multiple directions in parallel |
| Repetitive tasks | Plugins or manual work | One natural language prompt |
| Applying feedback | Read comments, edit manually | Summarize + auto-apply |
| Design system | Manual component lookup | @mention to reference directly |
| AI work location | Separate tab or app | Same canvas, same file |
The key shift is "same file." Every AI tool before this worked outside your design environment and handed you something to import. This agent is already in the room. Your team can comment on what it built, and it can iterate in response — a live feedback loop.
How to get started
- Join the waitlist
The agent is rolling out gradually in beta. Sign up for early access on figma.com. You'll need a Full seat on a Professional, Organization, or Enterprise plan. - Open the agent from the left rail
No plugin or separate setup required. The agent lives in Figma's left sidebar — open it and start prompting from any file. - Be specific in your prompts
"Three mobile onboarding screens with different visual directions" works better than "make some screens." Add @mentions for design system tokens and components to keep outputs on-brand. - Run agents in parallel
While one agent is generating concepts, assign another to clean up component naming or swap deprecated components. Running them simultaneously is where the real time savings come from. - Review as a team
Everything the agent creates is a regular Figma layer. Teammates can comment directly, and you can feed those comments back to the agent. No AI credits during beta — use it freely.
Where to start
Resist the urge to prompt "redesign our entire app." Start with something scoped: "dark mode version of this screen" or "swap all Button v1 instances to Button v2." Get a feel for how the agent interprets your design system before going bigger.
