AI Just Moved Onto The Design Canvas. Brilliant.design Is Betting Everything On It

Brilliant.design bets AI belongs inside the file, not beside it. It’s what GitHub and Figma combined might look like

Something real is shifting in design tools right now, and it’s not “AI got better at making mockups.” That part’s old news, tools have generated screens from a prompt for years.

What’s actually new in 2026: AI is moving into the workspace itself, the same jump coding agents made when they stopped just autocompleting lines and started operating inside a whole codebase, editing files, running things, fixing what broke. Design tooling is going through that same arc right now. A few pieces from this year have started calling it the Cursor moment for design, borrowing from how Cursor, the AI coding editor, broke through not with autocomplete but with AI working directly inside a real codebase instead of chatting alongside it.

A recent piece profiling how designers actually work in 2026 backs this up with real names, not just theory: designers running Claude Code and Cursor as part of their actual daily process, connected to Figma, GitLab, and other real tools through MCP. One line from it stuck with me: “the terminal has become a design studio.”

The shift, in plain terms:

• Old model: type a prompt, get a flat image or a rough draft back, then clean it up by hand yourself
• New model: the AI works inside the actual file, editing the specific layer or component you point it to, the way a teammate sitting next to you would

Brilliant.design’s Bet: GitHub and Figma Combined

I’ve tested Brilliant.design twice now, once on features, once on a follow-up test. Looking at it through this trend, Brilliant makes a lot more sense as a product decision.

Its files save as plain text, readable and diffable through git, the same way developers track code changes. Layer that onto a full vector design tool, auto layout, components, design tokens, the vocabulary any Figma user already knows, and you get something that behaves like GitHub and Figma combined into one tool. Its AI agent is built to edit inside that file directly, not bolt onto the canvas from outside.

To use the AI at all, you connect an agent, Claude Code, Cursor, Codex, or bring your own model through a provider like OpenRouter, DeepSeek and Gemini among the options there. In my testing, results varied a lot depending on which model was doing the work, a small free model broke layouts in ways a stronger one likely wouldn’t. That’s the real cost of this approach: power, but with setup and model choice standing between you and a working result.

Stitch’s Different Bet: The Clean Handoff

Google Stitch is making a different wager on the same shift. Instead of putting AI inside a new kind of file, it keeps the familiar prompt-to-screen model, but takes the handoff seriously: right-click a finished screen, export, and it lands in Figma with real auto layout and properly named layers, not a flat image or a pile of nested groups.

Alongside that, it also ships a DESIGN.md and MCP setup, plus a Generate menu with built-in accessibility checks.

Same underlying shift, both tools betting that AI should do more than generate a first draft, just at different stages of the process. Brilliant puts its effort into the building stage, letting the agent edit live inside your actual file as you work. Stitch puts its effort into the handoff stage, making sure what comes out lands cleanly in Figma instead of leaving you to rebuild it.

Even Figma’s own AI, Figma Agents, is part of this same movement, still labeled Beta, but clearly aimed at working inside the canvas rather than staying a separate chat window.

Where This Actually Leaves Things

None of these tools have this fully solved yet. Brilliant’s approach needs a connected agent and the right model to work well. Stitch’s strength is speed to a first draft, not deep in-file editing. Figma’s version is still Beta.

But the direction is the same across all of them: AI moving from just generating a picture to actually working inside your real files. Whichever tool gets there first is a genuinely open question. Worth watching either way.

I’m Usama, a UI/UX designer based in Islamabad. I write about what design tools actually do, not what they claim to do. More of these coming.

AI Just Moved Onto The Design Canvas. Brilliant.design Is Betting Everything On It was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.