The article provides a brief genealogy of how the concept of the rational agent was incorporated from economics into AI during the 1950s, through the framework of bounded rationality developed by Herbert Simon, operationalized in the so-called expert systems, reframed as a society of simple agents in the theory developed by Marvin Minsky, and later recognized as a central pillar of AI by the end of the twentieth century, until it was finally transformed into what I call a TaskBot—a hybrid human–machine assemblage designed to enhance specific productive functions. This genealogy shows that certain constitutive principles of AI are grounded in the division of labor and in the effort toward its rational organization. In this regard, the article examines the implications of procedural rationality, a concept aimed at reducing human cognition to the predictable and codifiable procedures that characterize the rational agent. With the rise of connectionist AI, this narrowing drive did not disappear, but was instead displaced onto data work—the human labor employed in the training of AI—and became a central design element in its management, directed toward the rational alignment of contemporary AI models.