Rough (energy, or free energy, or fitness) landscapes are ubiquitous in disordered systems, from material science to biology. These landscapes feature a multitude of local energy minima, separated by barriers and transition paths, with highly heterogeneous properties. In this talk, I will discuss how rough fitness landscapes emerge in the inference of statistical models for protein families. Such models can be extremely useful in understanding the emergence of protein structure and function during natural evolution, in generating new artificial proteins with desired properties, and in modeling in vitro evolution. I will discuss the validation of some of these models against two recent "in vitro evolution" experiments, and an ongoing experimental collaboration with the Tokuriki lab, to generate functional intermediates between natural antibiotic-resistance proteins.