Skip to content
IFAI

AI tools

Figma's new AI agent lets designers build plugins with prompts

Published Reach 53

Figma just introduced a design agent at Config that builds custom, reusable plugins and WebGPU shaders directly from text prompts. Designers can now spin up tailored tools on the canvas without dev setups or coding skills. It effectively turns prompt writing into functional workflow automation for teams.

Figma's new AI agent lets designers build plugins with prompts
Image: theverge.com

If you needed a custom tool in Figma to parse raw data or automate repetitive layer renaming, your options were pretty limited. You either searched the Community for a close-enough match or begged an engineer to whip up a script.

At Config on June 24, Figma unveiled its new design agent. The standout feature is simple but potent: you can now describe a custom workflow in plain text, and the agent builds a working, reusable plugin right on your canvas.

Turn prompts into tools right on the canvas

These generative plugins run entirely without developer setups, terminal windows, or coding chops. Because they tap into Figma's PropsKit, the resulting mini-apps look and feel just like native features.

Product designer Edward Chechique used the agent to build tools that import raw HTML into canvas elements and generate dashboard layouts. Instead of juggling external AI apps and copying code snippets back and forth, he now prompts the agent directly inside his open project.

It goes beyond basic utility scripts. The agent can also write custom WebGPU shaders on the fly, unlocking dynamic effects like dithering, pixelation, chromatic aberration, and procedural watercolor fills directly on Figma layers.

Creative technologist Anna Zhang treated the experience like a live negotiation, letting the agent build the UI controls while she dialed in complex photography treatments like marbling and light leaks.

Real context over generic text boxes

A prompt is only as smart as the material behind it. Figma addressed this by giving the agent access to project-level context.

You can drop user research docs, UX copy sheets, or analytics PDFs right into the chat. If you paste a link to an existing Figma file, the agent reads its full component hierarchy, design tokens, and style definitions rather than just looking at a flat image.

It can also perform live web searches and connect to external tools via MCP. That means mockups can pull in real restaurant names and photos instead of generic placeholder junk.

Running any generated plugin or shader is completely free across all plans, whether you made it yourself or borrowed it from a teammate. Prompting the agent to write new ones will eventually draw from paid AI credits once the feature reaches general availability.

Why it matters

Figma made a very smart tactical shift here. While the rest of the industry has been chasing text-to-UI generators that usually deliver messy, uneditable mockups, Figma is using AI to make tools, not just screens.

Every design team has idiosyncratic naming conventions, bespoke handoff routines, and tedious cleanup tasks. Giving non-technical designers the power to automate their own pet peeves in five minutes turns one-off ideas into reusable team assets.

For the broader industry, this resets the bar for creative software. The line between being a tool user and a tool builder is basically disappearing on modern canvases.

The catch? If a plugin needs to connect to third-party APIs or external AI endpoints, you still have to build classic plugins with traditional developer setups. And when an agent-generated script throws an unexpected logic error, non-technical designers will be entirely dependent on the agent to debug its own code.

Sources (2)

  1. [1]
    theverge.com(opens in a new tab)Originaltheverge.com
  2. [2]

Best of AI, daily