Google is quietly testing three experimental capabilities for its Gemini artificial-intelligence assistant, according to developer leaks and APK teardowns. “Agent Mode” is designed to let Gemini autonomously plan and execute multi-step tasks, while “Immersive View” would deliver richer, video-style answers, and a “Gemini Go” feature aims to support collaborative ideation, potentially tying into the company’s Canvas prototyping tools. The tests come as speculation mounts over a broader “Gemini 3” upgrade, but developers monitoring Google’s AI Studio say no preview release has appeared and expect the full model refresh to remain several months away. Google has not commented on timing for either the new modes or the next-generation model.