Proteins are dynamic, and this holds especially for their surfaces. They display ensembles of conformations, which allows them to interact with diverse partners, often via the same patch of surface, and execute their distinct functions. Binding a specific partner can stimulate — or suppress — a distinct signaling pathway. This diversity poses a challenge: how to reliably model a specific protein-protein interaction (PPI)? This problem is compounded in protein assemblies, which are typically large, involving multiple protein-protein interfaces. Integrative modeling (IM), which combines diverse data, has emerged as the most promising strategy; however, modeling dynamical interfaces, often at the detailed level, which are at the heart of reliable predictions of assemblies, still poses a challenge. Here we review hurdles and advances in integrative modeling of dynamical interfaces; while some could have been predicted or expected, others transformed modeling in unanticipated ways. We further comment on what we believe could be possible future advances.