For several years, our laboratory has worked on representing and using contextual knowledge to control real-world agents, for example, autonomous underwater vehicles (AUVs). Our approach, called context-mediated behavior (CMB), represents contexts as first-class objects called contextual schemas (c-schemas). Each represents a class of situation in which the agent may find itself, and it contains knowledge about how the agent should behave in that context. C-schemas are also instrumental in memory organization, forming the indexing structures of a content-addressable memory. Retrieval is augmented with a diagnostic process to find the appropriate set of c-schemas for a situation, which are then merged to create a coherent view of the context. Currently, we are extending our approach to handle more complex kinds of planning and acting and for use in multiagent systems (e.g., to provide context-appropriate ways to organize and reorganize the agents). This workshop paper gives an overview of the work, including the current status and future plans.