Classical AI focuses agent design purely on the agent. Various alternatives to classical AI, such as behavior-based AI and situated action, propose that agent design should include, not just the agent, but also its environment. These alternatives fall short of their own stated goals, however, in that they tend to include only the physical objects with which the agent interacts, leaving out such environmental factors as the audience with which the agent interacts, the people who are judging the agent as a scientific success or failure, and the designer of the agent him- or herself. Here, I argue that ignoring the social and cultural context of the agent actually leads to technical problems in the design of the agent, and propose a model for AI research that includes the full context of the agent. At the same time, I propose that the problems socially situated AI addresses are particularly pressing for applications in AI and entertainment, where the utility of an agent depends not so much on the internal correctness of the agent’s actions as on the effect of the agent on the user.