In nature, soft structures, such as protein assemblies, can organize reversibly into functional and often hierarchical architectures through noncovalent interactions. Encoding this dynamic capability in synthetic materials has remained an elusive goal. Through examples ranging from colloidal to molecular self-assembly I will illustrate the remarkable structures that have been realized experimentally in recent years. Specifically, I will discuss how computational modeling can contribute mechanistic understanding as well as predictive capabilities, despite the inevitable need for coarse-graining.