The scientific community’s increased exposure to cloud computing has led to increased familiarity with the machine virtualization technology that underpins the cloud. Efforts to define and implement provenance for the cloud are under way. In the meantime, however, an orthogonal idea, aimed at quickly facilitating repeatability and curation, has taken shape. This is the idea of using virtual machine images (VMIs) as authoritative, encapsulated and executable records of computations, especially computations whose results are destined for publication and/or re-use. In this paper we trace the origins of this idea, discuss its strengths and limitations, and outline some of the future steps and potential pitfalls on the path to its realization. We also analyze how this approach differs from, yet composes with traditionally defined provenance.