Recent research in artificial intelligence has developed computational theories of agents' involvements in their environments. Although inspired by a great diversity of formalisms and architectures, these research projects are unified by a common concern: using principled characterizations of agents' interactions with their environments to guide analysis of living agents and design of artificial ones. This article offers a conceptual framework for such theories, surveys several other fields of research that hold the potential for dialogue with these new computational projects, and summarizes the principal contributions of the articles in this special double volume. It also briefly describes a case study in these ideas—a computer program called Toast that acts as a short-order breakfast cook. Because its designers have discovered useful structures in the world it inhabits, Toast can employ an extremely simple mechanism to decide what to do next.