Descriptions:
Sam Witteveen demonstrates a significant update to Google Labs’ Stitch, a generative AI design tool that has evolved from a simple screenshot generator into a full agentic design platform. The update introduces a native design canvas powered by Gemini 3 Flash and Pro models, where users prompt the system to build and iterate on UI designs in real time—analogous to how coding agents work with an agents.md configuration file, Stitch now uses a design.md file that encodes brand colors, typography, icon styles, and component conventions.
Key new capabilities include URL ingestion for extracting design systems from existing websites, multi-page prototype generation with wired navigation, voice-driven design editing via Gemini Live, and direct export to AI Studio as a deployable Next.js application with authentication and database scaffolding. The tool also exposes an MCP interface for integration with external coding agents. Both the app and API tokens are currently free through Google Labs.
Witteveen walks through a practical demo—pulling a Thai resort website’s visual identity to generate a redesigned multi-page prototype—and compares Flash versus Pro model outputs, noting the Flash model can produce competitive results with faster iteration. For product designers, indie developers, and teams building early-stage web or mobile products, Stitch now offers a credible no-code path from visual concept to functional frontend prototype.
📺 Source: Sam Witteveen · Published March 19, 2026
🏷️ Format: Tutorial Demo
