Abstract Predictive breeding is now widely practised in crop improvement programmes and has accelerated selection response (i.e. the amount of genetic gain between breeding cycles) for complex traits. However, world food production needs to increase further to meet the demands of the growing human population. The prediction of complex traits with current methods can be inconsistent across different genetic, environmental and agronomic management contexts because the complex relationships between genomic and phenotypic variation are not well accounted for. Therefore, developing gene-to-phenotype network models for traits that integrate the knowledge of networks from systems biology, plant and crop physiology with population genomics has been proposed to close this gap in predictive modelling. Here, we develop a gene-to-phenotype network for shoot branching, a critical developmental pathway underpinning harvestable yield for many crop species, as a case study to explore the value of developing gene-to-phenotype networks to enhance understanding of selection responses. We observed that genetic canalization is an emergent property of the complex interactions among shoot branching gene-to-phenotype network components, leading to the accumulation of cryptic genetic variation, reduced selection responses and large variation in selection trajectories across populations. As genetic canalization is expected to be pervasive in traits, such as grain yield, that result from interactions among multiple genes, traits, environments and agronomic management practices, the need to model traits in crop improvement programmes as outcomes of gene-to-phenotype networks is highlighted as an emerging opportunity to advance our understanding of selection response and the efficiency of developing resilient crops for future climates.