In this paper I give a tentative assessment of the methodological status of agent based simulations. I first show under which conditions ABS can be a complement to traditional modelling. I then consider whether they can be held as a sound methodology of their own. Various topics relevant to the argument are briefly discussed, such as the forecasting role of theories, the realism of assumptions, Hayek’s insights on economics methodology. I cast the arguments given into some results of modern theory of abductive inference, providing a framework that helps clarify the logical status of simulations, and gives some hints on how to foster their role as a self sustained tool for economic reasoning.