introducing agentic video understanding with Gemini instead of using static processing (ingesting media at a fixed frame rate), the model can now take an active, goal-directed role in determining what to watch, at what speed, and through which modality (frames, audio, or