The architecture of technological and natural complex systems often exhibits hierarchical modularity.  Such architectures can be modeled with a directed and acyclic graph in which the inputs of the system appear at the bottom layer and the outcomes are produced at the highest layer. Such networks are often subject to an evolutionary process in which non-essential modules are removed.This evolutionary process often leads to an hourglass-like shape: starting from the bottom layer and moving up the hierarchy, the number of modules at each layer gradually decreases until the ``waist'' of the hourglass,  followed by a corresponding increase above the waist. The few modules at the waist have both high generality (i.e., they are reused in many higher layer functions) and high complexity (i.e., they depend on many lower layer functions). Further, these key modules at the waist are more persistent compared to modules in the input or output parts of the hourglass.  The talk will cover in more detail an instance of the hourglass effect in biological development, known as the genomic hourglass. This property of embryonic development predicts that divergent morphological patterns early and late in embryonic development are separated by a stage that shows the most morphologically similar patterns across different species. It has also been observed that the phylotypic stage expresses the oldest and most conserved transcriptome across different species. We have developed an abstract ``evo-devo'' network model that explains the emergence of the hourglass structure based on a topological property of the the underlying developmental gene regulatory network.