This (agentic object detection?) is very interesting. It looks like deductive recognition and detection vs. inductive training and learning from many samples. I need to look into it to understand how it really works under the hood, but on the surface it is certainly intriguing.