Google has introduced agentic video understanding in Gemini, extending the model from watching video to working with it.
From Watching to Working
Standard video understanding produces a one-shot summary or caption. The agentic version lets Gemini navigate long videos — find moments, compare segments, and answer follow-up questions grounded in what it sees. Google points at use cases from content review and moderation to sports analysis, where the value is in querying footage the way you’d query a document.
The feature ships as part of Google’s September wave alongside Gemini 3.8 Flash and Flash Cyber. Read the full announcement on the Google blog.