By kindergarten, our knowledge of agents has unfolded into a powerful intuitive theory that enables us to thrive in our social world. In this thesis I propose that children build their commonsense psychology around a basic assumption that agents choose goals and actions by quantifying, comparing, and maximizing utilities. This naïve utility calculus generalizes infants’ expectation that agents navigate efficiently, and captures much of the rich social reasoning we engage in from early childhood. I explore this theory in a series of experiments looking at children’s ability to infer costs and rewards given partial information, their reasoning about knowledgeable versus ignorant agents, and their reasoning about the moral status of agents. Moreover, a formal model of this theory, embedded in a Bayesian framework, predicts with quantitative accuracy how humans make cost and reward attributions.