In this paper, I motivate, define, and illustrate the notion of computationally grounded theories of agency. A theory of agency is said to be computationally grounded if we can give the theory an interpretation in terms of some concrete computational model. This requirement is essential if we are to claim that the theories we develop can be understood as expressing properties of real multiagent systems. After introducing and formally defining the concept of a computationally grounded theory of agency, I illustrate the idea with reference to VSK logic, a formalism for reasoning about agent systems that has a semantics defined with respect to an automata-like model of agents. VSK logic is an extension of modal epistemic logic, which allows us to represent what information is visible to an agent, what it sees, and what it knows. We are able to prove that formulae of VSK logic correspond directly to properties of agents.