Google unveils agentic design canvas
Google published a new agentic design tool that mixes a VR‑style canvas, voice agents, an 'Imagine new screen' prototype flow, and one‑click React code export with Figma/Studio integrations — the feature set was shared on social. The tool blurs designer→developer handoffs and could speed prototype→code cycles for full‑stack demos.
Stitch first shipped as a Google Labs experiment on May 20, 2025, and the launch post explicitly tied the project to Gemini 2.5 Pro for multimodal UI generation. (developers.googleblog.com) Google’s Canvas iteration runs on Gemini 3 and exposes a 1,000,000‑token context window to Google AI Pro/Ultra subscribers for large, multi‑screen projects. (gemini.google) Google AI Studio’s Build mode generates complete app scaffolds with React as the default client and a Node.js server runtime, and it surfaces the generated code in a live Code tab for direct editing or export. (ai.google.dev) Real‑time voice agents are supported via Gemini’s Live API and Google’s Agent Development Kit (ADK), with official guides showing streaming, low‑latency voice interactions and native audio models. (ai.google.dev) Google’s developer announcement documents a “paste to Figma” flow and front‑end code export, while hands‑on reviews from LogRocket and other testers report that exported code often needs manual cleanup and that some export paths vary by model mode. (developers.googleblog.com) The public Stitch/Gemini ecosystem already includes a Stitch skills repository and Agent Skills integrations on GitHub, showing Google’s intent to standardize agentic plugins and connect to external tooling like Figma, AI Studio, and third‑party code editors. (github.com)
