At any given point in time, an agent's knowledge of the external world is her perceivable environment and a map of the world that she constructs as she goes along. A plan is created for a given goal, in a given context, by either selecting an old plan from memory (i.e., a previous execution trace) or creating a new plan from scratch. Old plans are stored somewhat abstractly so they require some refinement before they are deployed. In either case, agents are assumed to be adaptive planners (Alterman, 1988), so if the plan does not exactly match the current circumstances, differences and changes will be teased out during the ongoing interaction between the agent and the environment. Much of the adaptation of plans (both new and old) that occurs results from communication between agents at runtime.
An agent creates a plan from scratch using a given set of STRIPS-like operators (Fikes & Nilsson, 1971) using a hierarchical search strategy like ABSTRIPS (Sacerdoti,1974). Unlike other distributed planners (e.g., Corkill, 1979), the agents do not use communication in their planning process. This means that the plans generated may not be globally efficient or even feasible. Depending on their type, different agents have different operations that they can perform.