Keith Decker, Katia Sycara and Mike Williamson
One of the most important uses of agent models is for problem-solving coordination. Coordination has been defined as managing the interdependencies between activities, or pragmatically as choosing, ordering, and locating actions in time in an attempt to maximize a possibly changing set of decision criteria. Coordination activities include not only localized agent interactions over specific problems, but also longerterm agent organizations that can support current and future problem-solving activity. Models that support coordination can be divided into three classes: models of other agents’ current intended actions, schedules, and/or plans; models of other agents’ objectives (desires, goals); and models of other agents’ capabilities. This paper will focus on the longer-term models of capabilities that agents need in order to organize effectively to solve problems. Such models deal with an agent’s capabilities and long-term commitments to certain classes of actions. Besides discussing the models themselves, we will discuss the basic behaviors agents should use to construct, communicate, and update these models. Finally, we will discuss how these models can support several different organizational forms such as economic markets, Wiederholdian federations, and bureaucratic functional units. The work will be presented in the context of an implemented agent architecture for Internet information gathering in dynamic domains such as financial portfolio management (called WARREN).