In this position paper, we present an overview of recent work on using simulation to evaluate agent architectures. We characterise the kinds of evaluation that can be performed using simulation, and highlight some of the key issues in using simulation for evaluation. Two main trends in recent work are then identified: an increasing adoption of ideas from the `mainstream' simulation community (particularly parallel discrete event simulation), and an increasing focus on the interoperability of simulators. We characterise the features that distinguish MAS simulation classical parallel discrete event simulation problems and highlight recent work which attempts to address these problems. We argue that, together, these developments point to the emergence of new software engineering methodologies for MAS, in which simulation (and hence evaluation) plays a central role throughout the development process.