Recent advances in computational protein design have depended mainly on neural networks and machine learning to generate binders. However, the complexity of protein–protein interactions and the limitations of data-driven models constrain future progress.