Multi-agent systems can be considered from a variety of perspectives. One such perspective arises from considering the architecture of an agent itself. Another is that of an instantiated agent architecture and its interaction with its peers in a MAS. A third perspective is that of an external observer. These three perspectives cover a potentially overlapping but essentially distinct set of issues concerning MAS simulation and modelling. In this paper, we consider each of these perspectives in turn and demonstrate how a simulation framework can support a collective treatment of such concepts. We discuss the implications for agent development and agent society design arising from the results and analysis of our simulation approach.