For years, product teams have talked about “getting everyone in the room” to jam on ideas together. With Google Stitch’s latest update, that room has basically moved into the browser – and the “everyone” now includes an AI agent that can sketch, revise, document, and even ship code alongside you in real time.

Stitch started out as a clever Google Labs experiment: type a prompt like “design a mobile dashboard for tracking orders” and it would spit out a surprisingly polished interface plus front-end code. At Google I/O, Google pushed that idea much further, turning Stitch into what it calls an AI-native design canvas where design, copy, and production code evolve together while a Stitch Agent streams changes directly on the screen. The pitch is straightforward but ambitious: instant team creativity. Instead of designers, developers, and PMs bouncing Figma links and spec docs back and forth for weeks, everyone can talk to the same canvas – with an AI agent acting as the world’s fastest, infinitely patient design collaborator.

In practical terms, the new Stitch feels less like a traditional tool and more like a live design session that never really ends. You can start with something as simple as a text description, a voice note, or even a screenshot or rough wireframe, and the system turns that into a fully formed UI for web or mobile. Under the hood it leans on Google’s latest Gemini models to interpret what you’re asking for, apply layout and UX conventions, and then fill in the gaps with components, colors, and copy that actually make sense together. What’s new is how you work with that output: instead of waiting for a static design to render, Stitch now streams its work to the canvas in real time, so you can watch the layout evolve and course-correct mid-generation if it starts drifting away from your vision.

That streaming behavior changes the rhythm of design. If you’ve used generative design tools before, you’re probably used to the loop: type a prompt, wait for a few options, reject most of them, then tweak the prompt and try again. Stitch’s real-time mode turns that into more of a conversation. You might see an initial layout form and immediately say, “Make the hero section taller and swap the illustration for a product screenshot,” or “Dial the overall style back – this is a financial dashboard, not a festival poster.” The agent responds on the canvas while you talk, almost like a designer sharing their screen in a video call and making edits as stakeholders speak up.

This is where the “team” part shows up. Google has been framing Stitch as “vibe design” – their term for an AI-native workflow where ideas move fast, feedback is conversational, and the boundary between design and implementation gets blurry. In practice, it means multiple people can join a Stitch project, see the same canvas, and nudge it from different angles, while the agent mediates and keeps track of what’s changed. A PM might be focused on messaging clarity, a designer on hierarchy and spacing, and a developer on whether the layout can actually map to real components; the agent sits in the middle, listening and translating that pile of opinions into concrete adjustments and, importantly, production-friendly code.

That bridge to code is a big part of the story. Earlier versions of Stitch were already able to export front-end code, turning designs into HTML, CSS, or framework components in minutes. Now Google is leaning into that pipeline with “skills” – modular abilities that let the Stitch Agent generate design documentation, React components, design tokens, and other developer-facing artifacts directly from the same source of truth. With the Design MD skill, for example, the agent can scan your Stitch project and spit out a structured design spec – colors, typography, spacing rules – that a developer can drop into a repo as DESIGN.md. Another skill focuses on React components, turning screens into production-ready code that fits into existing workflows rather than sitting as pretty but inert mockups.

Unsurprisingly, Google is tying Stitch into its wider developer story. In a recent codelab, Google walks through connecting Stitch to its Antigravity IDE via an API key, letting an agent pull design metadata, generate documentation, and then build a working React and Tailwind project from the same design. The flow is almost sci-fi: you describe a SaaS landing page in Stitch, refine the layout on the canvas with the agent, then switch over to Antigravity and ask another agent to “fetch the LaunchPad design and build the full website.” The agent pulls palette and typography from Stitch, writes the components, starts a dev server, and even adjusts specific visual details – like button padding – by referring back to the original design context when you complain something looks off.

All of this adds up to a subtle but important shift in what “design tools” even are. Stitch is not just a canvas with smarter auto-layout; it’s closer to an AI coworker embedded inside the canvas. You can speak to it with voice commands, ask for critique, and have it track the evolution of a design over time. If you’re not a designer, that’s empowering: instead of struggling through a complex interface or learning design jargon, you can say “make this feel more minimal and enterprise-friendly” and let the agent explore options on your behalf. If you are a designer, it’s less about replacing your taste and more about having a pair of hands that can churn through variations while you focus on direction and nuance.

There is, of course, the looming question of what this means for the design tools people already rely on every day. Figma has been steadily adding AI features of its own, and a cottage industry of AI-assisted design platforms is emerging around similar promises: faster iteration, automatic layout, instant copy tweaks. Where Stitch tries to stand out is in how deeply AI is baked into the workflow from the start. It assumes the primary way you interact with design is through language and examples – not through drawing rectangles and manually wiring components – and then builds everything else around that assumption. Google’s play is that an AI-native tool built directly atop Gemini, with hooks into Google Cloud, IDEs, and other agents, will feel more like a collaborative environment and less like a design file editor with AI bolted on.

If you zoom out a little, what Stitch is really testing is how comfortable teams are with letting AI sit in the middle of their creative process. On one hand, it promises to smooth out all the tedious parts: aligning components, rewriting microcopy yet again, updating specs after every tweak, regenerating code when a button moves ten pixels. On the other, it challenges some of the rituals that design and product teams have used for years – the pixel-perfect handoff, the carefully managed design system, the guarded Figma file. In a Stitch-style world, a lot of that structure is reconstructed on the fly, guided by prompts and conversation instead of rigid processes.

It’s also not hard to imagine the cultural tension this creates. For small teams and startups, the allure of going from “rough idea in a meeting” to “interactive prototype with working code” in a single afternoon is obvious. For larger organizations, where design ops, accessibility, and brand governance matter, the question will be how well Stitch’s agent respects the rules – whether it can be trained on existing design systems, enforce constraints, and prevent a slow slide into AI-generated chaos. Google’s emphasis on skills, design tokens, and structured docs suggests it sees Stitch as something that should plug into those systems, not blow them up.

Still, the phrase that keeps coming up around Stitch – “instant team creativity” – is less about speed for its own sake and more about who gets to participate. When the canvas listens to voice, accepts sketches and screenshots, and can explain its own decisions in natural language, you lower the barrier for people who aren’t formally trained in design or front-end dev to meaningfully shape a product. That doesn’t eliminate the need for specialists, but it changes when they enter the process and what they spend their time on. Instead of being bottlenecks for every small experiment, designers can curate, critique, and refine the wave of ideas that teams generate together with the agent.

Google is still calling Stitch an experiment, and there are plenty of open questions: how well it behaves on messy, real-world projects, how teams integrate it with existing tools, and how comfortable people are with AI being present in every step of their design conversations. But with real-time streaming, agent skills, and deep links into developer workflows, Stitch is no longer just a flashy demo from I/O. It’s an early look at a future where the “design meeting” is something you can start from your browser, invite your team to, and never really end – because the canvas, and the AI inside it, are always ready to pick up the thread of the next idea.