Amazon's New AI Watches Your Videos and Answers Questions

Amazon Web Services published a blueprint for an AI system that turns video into something you can simply chat with. You upload a 90-minute meeting, a day of security camera footage, or hours of field inspection clips, then type questions in everyday language: "What decisions were made?" or "Which car changed lanes before the crash?" The system answers in seconds. No more scrubbing through hours of footage hoping to spot the right moment.

The clever part is how it decides what to do. Until now, each type of question needed its own custom-built pipeline of code — one for transcription, another for spotting faces, another for visual searches. This system uses an "agent" (AI that picks its own tools and takes actions) that chooses at runtime whether to transcribe speech, analyze visuals, or reuse an earlier answer. If the video was already processed, you get an answer in under a second. Brand-new video takes 5–10 minutes to analyze, depending on length.

A major media and entertainment company tried it during an AWS consulting project. Their staff queried recorded discovery sessions to pull out design decisions, action items, and who said what. According to the company's own internal comparison, manual review time dropped about 80% across a backlog of more than 200 multi-hour recordings. That figure comes from the customer, not an independent study, so treat it as promising rather than proven.

Why anyone else should care: media teams, security departments, insurers, and inspectors all sit on piles of video nobody has time to watch. The catch is real, though. This isn't a consumer app — you need an AWS account, coding know-how, and paid cloud services. Setting it up takes technical work, costs aren't published, and face-matching across footage raises obvious privacy questions about who gets watched.