Agent based models (ABMs) are a useful tool for modeling population dynamics, where many details can be included in the model description. The computational cost though is very high and for stochastic ABMs a lot of simulations are required to sample quantities of interest. Especially for high agent numbers the sampling is often infeasible. In this talk we will discuss possible model reduction steps that lead to a significant gain in compuational efficiency while preserving important dynamical properties, such that the statistics for first hitting time events are still similar. In particular we will consider diffusive ABM, metapopulation and piecewise deterministic approaches for population dynamics and highlight how they are connected. We will illustrate our coarse graining concept by applying it to an ABM for spreading processes.