In this paper, we present the concepts behind die implementation of behavioral and cognitive capabilities in objects used for controlling workstations in manufacturing systems. These capabilities characterize a new class of objects called agents; such objects intuitively play die role of human decision-makers in the system modeled. Three types of reasoning activities are considered: a meta-reasoning level to control the inference/search reasoning activities, die local reasoning level incorporating simple rule-based procedures and an extended reasoning level representing the reasoning activities that cannot be performed using simple procedures and thus require external knowledge agents to assist the decision-making process. Implementation issues are discussed, illustrated by an application of the approach to the modeling of workstations in a rolling-mill facility.