Two different reasons for using agents are distinguished: the `engineering' perspective and the `social simulation' perspective. It is argued that this entails some differences in approach. In particular the former will want to prevent unpredictable emergent features of their agent populations whilst the later will want to use simulation to study precisely this phenomena. A concept of `social embeddedness' is explicated which neatly distinguishes the two approaches. It is argued that such embedding in a society is an essential feature of (...) being a truly social agent. This has the consequence that such agents will not sit well within an `engineering' methodology. (shrink)