Google is trialing three experimental modes for its Gemini generative-AI assistant—Agent Mode, Gemini Go and Immersive View—according to information circulating among developers. Agent Mode is designed to plan and complete multi-step tasks autonomously, Gemini Go focuses on collaborative ideation and rapid prototyping, while Immersive View produces visual or video-style responses that extend the system’s existing “Video Overviews” capability. The tests follow a series of Gemini upgrades announced or rolled out in recent days. Users with paid Workspace plans can now question Gemini about images stored in Google Drive, and a new memory system lets the assistant personalize answers while a “ghost” option deletes individual conversations. Google has also introduced the Gemini 2.5 Flash image model to speed up image generation. No release timetable has been provided, but the expanding feature set underscores Google’s strategy of embedding Gemini deeper into Workspace and consumer services as competition with OpenAI, Microsoft and Anthropic intensifies.