Thank you, Daniel, for the feedback and the links. I have already been aware of the works of Charles Macal and the little book by Gaylord is nicely sitting on a shelf nearby.

My question is not as open-ended as it may seem: I did very precisely ask about the ability of Wolfram System Modeler to address agent-based models, e.g. where structure is created during run time and also terminated again. Also there is the ability to address multi-paradigm modeling (ABM, System Dynamics and Discrete Event Simulation) like AnyLogic further down the line of my question.

I am aware of the potential of Mathematica to address ABM especially when linked to other software like REPAST. But that is the point: Writing equations for larger models -- especially with structure that is repeatedly used -- is imho not really an option, so one needs some kind of declarative programming tool where graphical building blocks can be used (and reused).

So far it seems that modeling variable structure systems is not possible with WSM?

Since I have not received any reply here it seems that either nobody is able to answer my question or nobody is interested. In comparing AnyLogic with WSM I have formed a hunch that the former will very likely be the better tool of choice when agent-based modeling in the sense of variable structure modeling is of importance.

One might probably come to the somewhat simplified solution that if managerial problems of system control (after all: what is management about?) are of interest a modeler might rather turn to AnyLogic than to WSM. The sterotypcial engineering problem on the other hand seems nicely addressed by Modelica.