Research editor
A world model learns what a scene will do next. From Ha and Schmidhuber in 2018 to V-JEPA, Genie, Sora and Cosmos, the research is moving toward video. What that could mean for a camera over a loading dock, and which parts are still claims with no result behind them.
SAM turned segmentation into a prompt. SAM 2 gave it memory across frames. Put a self-supervised embedding behind them and a site manager can define a new object class from a handful of clicks, with no retraining. The design, the papers, and the limits.
Describing a scene is a solved demo. Doing the right thing in it every hour of a shift is the harder test, and buyers have started writing four metrics for it into their evaluations.
In a trends piece for SourceSecurity, Hanwha's John Lutz Boorman argues that AI agents will take over routine watching and initial response, leaving humans with judgement calls and an oversight role the company calls AI governance manager.
Analyst forecasts now put agentic capability in half of all video analytics by 2030, up from under a tenth today. A latency budget and a utility function explain why the curve bends where it does.
A forklift, a pallet and a person, and the moment one hides another. A video agent is judged on what it does in that second, and the sales deck skips it.
The models that win benchmarks assume a data-center GPU. The box behind a loading-dock camera has a 15-watt budget. Distillation, quantisation, and pruning are how one becomes the other, and a latency budget decides what stays behind.
Engineering News-Record's Bryan Gottlieb found carriers cutting water-damage deductibles in half for sites with leak sensors. The mechanism matters for video agents: underwriters have started pricing continuous monitoring, and video is a form of it.
A vision-language model is two encoders and a shared vector space. Feed it video and the hard part becomes time, and asking it to act takes a different model again. The papers, in order, from ImageNet to RT-2.
Carriers are starting to give credit for continuous monitoring on construction sites, in fleets and in workers' comp. None of them has yet written a rate for a video agent, and more than 60 of them are filing to exclude AI from liability cover.
Spot AI stacked small detection models together until the stack ran out of judgement. A large multimodal model closed the gap, and the agent category came out of the seam between the two.
Insurance Business's Chris Davis reports on how the carrier uses telematics, sensors, wearables and camera AI to prevent losses before they become claims, and why its risk engineers see productivity gains alongside safety.
Text sits still. The physical world moves, and it punishes slow answers. Four requirements that make video agents a different engineering problem from chatbots.
Interviewed at RISKWORLD 2026 by Risk & Insurance, Matt Scott described AI watching workplace video for hot spots, reckless forklifts and badly packed pallets. Discounts, for now, are for camera-equipped fleet telematics.
Retrieval over video borrows its shape from text RAG, then has to solve two problems text never had: where in the frame, and when in the stream. The research that makes it work, and the checking step that keeps false positives down.