Agentic Object Detection: Finding Objects From a Text Prompt, No Labels

Agentic Object Detection lets a user specify what to locate in an image with a short description. Andrew Ng's announcement gives unripe strawberries as an example and says users need not label training examples for the requested task.

LandingAI's technical explanation, published later that month, describes planning, code generation and tool use to reason about the image. The output is a set of boxes locating detected objects. That makes the capability useful for requests combining an object with an attribute, rather than merely asking for a general image description. This describes the vendor's supported interface and intended use; it does not establish accuracy on a reader's own images.