Google Stitch launches real-time AI agent and multiplayer editing at I/O 2026

Google updated Stitch at I/O 2026 with two capabilities that shift it from an experimental design tool into a more serious competitor for UI design workflows: real-time streaming design generation and simultaneous multiplayer editing.

The streaming model replaces the previous turn-based interaction. Rather than submitting a prompt and waiting for a complete output, the Stitch Agent now renders UI components directly to the canvas as a designer types or speaks. Layouts reflow in real time, and designers can redirect the agent mid-generation — stopping a direction that is heading the wrong way before it finishes. This is a meaningful shift: it moves the design experience closer to working with a collaborator who responds immediately, rather than a tool that produces outputs for approval in discrete rounds.

Multiplayer editing lands through a new Agent Manager layer. This interface logs the full evolution of a project and helps teams explore multiple design directions in parallel. Finished designs can be shared instantly via a link through Google AI Studio, exported to Google Antigravity for backend integration, or published directly to the web via Netlify. The MCP server and SDK expose Stitch’s capabilities as machine-callable tools, enabling teams to embed Stitch into CI pipelines and third-party agents.

Stitch accepts multi-modal input: text prompts, voice commands, and imported codebases or design files. It runs on the Gemini model family and is currently free through Google Labs with Standard and Experimental generation tiers.

The comparison to Figma is direct. Figma’s recent releases — including Figma Make, Weave, and the May 2026 release notes covering agent workflows — show the same direction: AI generation inside the design canvas with iterative control. Figma charges $15 per seat for its AI features. Stitch is free for now.

For product teams and UI designers, these tools represent the same underlying shift: design is becoming less about placing elements manually and more about directing an agent that handles the mechanics while the designer focuses on decision-making and evaluation.