DeepMind has unveiled a generative browser concept powered by Gemini 3.1 Flash-Lite.

The prototype, which operates within AI Studio, generates HTML and CSS on-the-fly as the user browses, rather than loading pre-rendered pages from servers. Both the visual layout and content are dynamically shaped by user prompts, clicks, and the overall navigational context.

This technology fits naturally into the framework of autonomous AI agents, which may need to instantaneously assemble temporary dashboards or utility tools to complete a specific task.

However, the project is still far from polished. Since both the structure and content are generated by a lightweight version of Gemini, the browser is prone to vulnerabilities, minor hallucinations, and layout distortions. Furthermore, the system remains highly resource-intensive, consuming a massive amount of tokens for inference.