We present a progress report on our efforts to establish a new research program for evolutionary systems biology, based on reverse-engineering and in silico evolution. The aim is a mechanistic understanding of the genotype-phenotype map and its evolution. Our review focuses on the case study of the gap gene network in dipteran insects (flies and midges). This network is the top regulatory tier of the segmentation gene hierarchy, generating a pattern of overlapping expression domains that subdivide the embryo during early embryogenesis. It is one of the best-understood developmental regulatory networks today. We have studied this system in a comparative way, across three species: the vinegar fly, Drosophila melanogaster, the scuttle fly, Megaselia abdita, and the moth midge, Clogmia albipunctata. In this context, we discuss methodological challenges concerning data processing and model-fitting, consider different functional decompositions of the gap gene network, and highlight novel insights into network evolution by compensatory developmental system drift. Finally, we discuss the prospect of simulating the phylogenesis of the gap gene network using in silico evolution. We conclude by arguing that our case study is a first step towards a more systematic empirical investigation into the principles of network evolution.