Cosmological data from a multitude of probes has been pouring in at unprecedented rates over the last decade or so, with a corresponding dramatic rise in statistical precision. Precision is not the same as accuracy, however, and we now face the challenge of beating down to acceptable levels the wall of systematic uncertainties in our models of the Universe, particularly in the study of Large Scale Structure. A popular (and very powerful) way to address this problem is to run large, high-resolution simulations of cosmological volumes and "calibrate away" uncertainties. Apart from computational resources, running and analysing simulations requires its own unique skill sets, and the question arises: Is there any room left for approximate analytical techniques in this demanding environment? I will argue using two examples -- the size distribution of HII bubbles during reionisation and cosmological constraints from galaxy clusters -- that although numerical skills are now mandatory, the days of the blackboard cosmologist are not quite done yet. E.g., I will show that naive applications of numerical fits to simulations can lead to significant biases when recovering cosmological parameters in surveys such as Planck, and can incorrectly point towards new physics. Meeting the challenge of the "systematics wall" will therefore require a combination of numerical and analytical efforts.