Vibe Design Is Real: Inside Google Stitch's March 2026 Update
Google's AI design canvas just got an infinite workspace, a design agent, voice interaction, and a portable design system format. Here's what it means for designers.
Two days ago, Google Labs product manager Rustin Banks published a post introducing what he calls “vibe design”, a term that’s either going to define a new era of design tooling or become the most mocked phrase at every design conference this year. Probably both.
The tool is Stitch. It launched quietly at Google I/O last year as a text-to-UI generator, impressive party trick, limited practical value. What Google just shipped is something fundamentally different: an AI-native design canvas with an infinite workspace, a design agent that reasons across your entire project, voice interaction, instant prototyping, and a design system format called DESIGN.md that bridges design and code tools through an agent-readable markdown file.
Let’s unpack what actually matters here and what it means for anyone who designs interfaces for a living.
“Vibe Design” Is a Terrible Name for a Real Idea
I’ll say it: “vibe design” sounds like something a SaaS founder says at a WeWork happy hour after two Aperol Spritz. But beneath the cringe, there’s an insight worth taking seriously.
The traditional design process starts with structure. Open Figma. Set up your grid. Choose your breakpoints. Pick a type scale. Place a rectangle. You’re making hundreds of micro-decisions before you’ve answered the most important question: what should this thing feel like?
Stitch flips the sequence. You start by describing the business objective, the emotional register, the references that inspire you. The AI generates multiple high-fidelity UI directions, not wireframes, not mood boards, actual screens, and you diverge before you converge. It’s brainstorming at the speed of rendering.
This isn’t new thinking. Anyone who’s run a proper design sprint knows the value of starting wide and narrowing late. The difference is that Stitch collapses the time between “let’s explore” and “here are five directions to react to” from days to minutes. The diverge-converge loop that takes a design team a week now happens in an afternoon. Alone.
That’s the disruption. Not the AI. The compression.
The Canvas Is the Real Story
Everyone’s going to write about voice input and vibe design because those are the headline features. But the most significant shift is the canvas itself.
Stitch’s new infinite canvas isn’t a Figma clone with AI bolted on. It’s an AI-native workspace where you can throw in anything, screenshots, text descriptions, code snippets, competitor references, a paragraph describing your user, and the design agent uses all of it as simultaneous context. The mental model isn’t “tool with AI features.” It’s “AI with a spatial interface.”
This matters because it changes the unit of design input. In traditional tools, the input is a precisely defined action: draw a rectangle, set a color value, align these layers. In Stitch, the input is intent. You’re not instructing a tool. You’re briefing an agent.
I’ve been calling this shift the move from Interfaces to Experiences to Agent Design for two years now. Google just built a product that embodies it.
DESIGN.md: The Feature Nobody’s Talking About
Here’s where it gets interesting for anyone running a design system at scale.
Stitch introduced DESIGN.md, an agent-friendly markdown file that captures your design rules (colors, typography, spacing, components) in a format that can be exported from one project and imported into another. Extract a design system from any URL. Port it across Stitch projects. Feed it into coding tools.
If this sounds familiar, it should. It’s the same problem I’ve been working on with a client, evolving traditional design systems into what I call Digital Experience Systems (DXS). The challenge has always been that design tokens live in Figma, code tokens live in the codebase, and the translation layer between them is a manual, fragile process that breaks every time someone changes a variable.
DESIGN.md is Google’s answer to that translation layer. And the format choice, markdown, is deliberate. Markdown is the lingua franca of AI agents. Claude reads it. Gemini reads it. Every coding assistant on the market reads it. By choosing markdown over JSON or proprietary formats, Google made design systems portable across the entire AI toolchain.
This is the same instinct behind the CLAUDE.md files I’ve been writing for my own projects — giving AI agents the context they need to make design-aware decisions. The difference is that Stitch generates the file from your actual designs, closing the loop between visual output and machine-readable input.
For design leaders: pay attention. The design system of 2026 isn’t a Figma library with documentation. It’s a DESIGN.md file that travels between your design agent, your coding agent, and your prototyping environment. If your system can’t be read by a machine, it’s already legacy.
The Agent Manager and Multi-Direction Exploration
Stitch also ships with what Google calls an “Agent Manager”, a feature that lets you work on multiple design directions in parallel while the agent tracks progress across all of them.
This is quietly radical. Traditional design tools are single-threaded by nature. You work on one direction at a time, and switching between alternatives means duplicating frames, managing naming conventions, and losing context. The Agent Manager treats parallel exploration as a first-class workflow. You can have the agent pursuing three different visual directions simultaneously, each building on the shared project context.
For anyone who’s ever presented design options to a stakeholder and watched them pick the safe one because you only had time to polish two directions, this is the fix. When the cost of exploration drops to near zero, the quality of the final output goes up because you can afford to explore the weird, ambitious, “probably won’t work but let’s see” direction alongside the safe bet.
Voice: The Real Collaboration Is the Conversation
Google added voice interaction to Stitch. You can speak to your canvas, and the design agent will critique your work, generate variations, ask clarifying questions, and make real-time updates as you talk.
I’ll be honest, my first reaction was skepticism. Voice interfaces in creative tools have historically been gimmicks. But there’s something different about using voice during the early exploration phase of design, when you’re not yet ready to commit to specific decisions. Typing precise prompts feels too deliberate when you’re still vibing (there, I said it). Speaking your thoughts feels closer to how designers actually think in the early stages, loose, associative, contradictory.
The more interesting implication is the conversational critique. Having an AI that can look at your canvas and say “this layout creates visual tension between the hero and the CTA” or “your color choices feel clinical for a wellness brand” is something designers usually only get from a senior design review. Stitch is offering design critique as a feature. That’s a statement.
MCP Integration: The Bridge That Matters
Stitch now ships with an MCP server and SDK, meaning you can connect it to coding tools like Gemini CLI, Claude Code, Cursor, or Antigravity. The Stitch Skills library on GitHub already has 2,400+ stars.
This is where Google’s play gets strategic. By opening Stitch to the Model Context Protocol, they’re positioning it not as a destination tool but as a node in a larger agentic workflow. Design in Stitch, export via MCP, build in your coding agent of choice. The design system travels with DESIGN.md. The context travels with the agent.
For the workflow I’ve been building — Figma for production design systems, Claude Code for implementation, GSD for orchestration — Stitch could slot in as the rapid exploration layer. The place where you go from brief to concept before anything hits Figma. Not replacing the production workflow, but compressing the front end of it.
What This Means for Designers (The Honest Version)
Let me be direct: Stitch is not going to replace Figma. Not this year, probably not next year. Production design systems, team-scale collaboration, pixel-perfect refinement, accessibility auditing — Figma is still where that work lives. Google knows this. The practical workflow they’re implicitly suggesting is: explore in Stitch, refine in Figma, build with AI coding tools.
But here’s the uncomfortable question Stitch forces: if a non-designer can go from business objective to clickable high-fidelity prototype in an afternoon, what exactly is a designer’s value?
The answer isn’t “better pixels.” It’s never been about better pixels. The answer is judgment. Knowing which of the five AI-generated directions actually solves the user’s problem. Understanding that the beautiful layout fails on accessibility. Recognizing that the “calm and minimal” direction the stakeholder loves will tank conversion because it buries the primary action. Holding the tension between what looks good and what works.
This is the argument I keep making: in an age of infinite generation, curation becomes craft. Stitch gives everyone the ability to generate. It doesn’t give them the ability to evaluate. That’s us. That’s always been us.
But only if we stop defining our value as “the person who uses the design tool” and start defining it as “the person who knows what good looks like and why.”
The Bottom Line
Google Stitch’s March 2026 update is the most significant development in AI-native design tooling since, well, the first version of Stitch. The infinite canvas, DESIGN.md, the agent manager, and MCP integration aren’t just features — they’re an architecture for how design work will happen in the age of agents.
The “vibe design” branding will get mocked. The tool has real limitations, no deep prototyping with animations, no design system integrations at Figma’s depth, monthly generation limits. But the direction is unmistakable: design is moving from manual construction to intelligent orchestration.
And for those of us who’ve been saying that for a while, it’s nice to see Google agree.
Try it yourself: stitch.withgoogle.com
Great overview!