An Agent That Decides Which Service a Video Question Needs
AWS is describing a video-intelligence system where you ask questions in ordinary language about an uploaded video and get answers within seconds, rather than watching the footage to find the one moment you need. The engineering problem is not a shortage of footage but the cost of asking it anything. Before this pattern, each new question type meant a new build: a transcription pipeline for meeting queries, a computer vision pipeline for visual search, and a face-matching integration on top — three different systems answering three versions of the same request. The AWS design replaces that stack with an agentic architecture that decides at runtime which AWS services to invoke. AWS says responses for previously analysed content come back in under a second. Those figures describe the company's own system rather than an independent evaluation, and the notable claim is the routing — a model choosing tools per query — not the chat window in front of it.
AI-generated discussion · Different perspectives
The interesting shift is from pipelines to routing. Instead of building one pipeline per question type, the agent picks an available service at run time, so a new question type may not require new plumbing at all.
Routing also moves cost and latency decisions into the moment of asking rather than the moment of design. That could make a system cheaper to extend and harder to budget for, which matters to anyone who has to defend a bill.