transport

Proposes two-parameter flows to learn the dynamics of high-dimensional probability densities from unlabeled samples, using conditional flow matching to extract physics-time velocity fields.

The article argues that AI agents are shifting from synchronous chat interfaces to asynchronous background workflows, highlighting new features from Anthropic, OpenAI, and Cursor that decouple agent lifetimes from HTTP request-response cycles.