How we choose to represent phenomena of interest in the form of models has important implications for the inferences we can sensibly draw from the models and apply back to the phenomena of interest. I will discuss how common approaches for longitudinal modelling may be inadequate for scientific inference, argue that stochastic differential equations may provide a more appropriate basis for such inference, and present the ctsem software for hierarchical continuous time state-space modelling. Along the way I will show how complex forms of heterogeneity across subjects and time may be handled, discuss the modelling of developmental phenomena, the inclusion of intervention effects in a systems model, and present a likely naive but hopeful viewpoint for directions that a serious approach to dynamic systems thinking could take us.