Large-scale machine learning has given rise to agentic AI systems, which the paper describes as persistent, autonomous computational entities that pursue goals, initiate actions, and interact with open environments. These systems differ from traditional software in that they operate continuously and with a degree of independence.

The authors argue that this emerging class of systems calls for a new organizing abstraction—agency—to complement existing abstractions in computer science. The abstract suggests that agency could serve as a foundational concept for reasoning about such systems, though the full argument is not visible in the provided text.

Treating agency as a first-class abstraction could reshape how computer scientists design, analyze, and teach about AI. The specifics of the proposal remain to be examined, but the framing points to a shift in how the field conceptualizes autonomous computation.