I describe an evolutionary procedure in silico that creates small gene networks performing basic tasks. I use it to evolve a wide range of models for very different biological functions: multistability, adaptive networks and entire developmental programmes like somitogenesis and Hox gene pattern. In silico evolution finds both known and original network designs, and can be used to make predictions on biological behaviours. This computation illustrates how complex traits can evolve in an incremental way, and suggests that dynamical systems theory could be used to get new insights towards a predictive evolutionary theory.