In this paper we propose a theoretical framework to understand the governance of internet-mediated social production. Focusing on one of the most popular websites and reference tools, Wikipedia, we undertake an exploratory theoretical analysis to clarify the structure and mechanisms driving the endogenous change of a large-scale social production system. We argue that the popular transactions costs approach underpinning many of the analyses is an insufficient framework for unpacking the evolutionary character of governance. The evolution of Wikipedia and its shifting modes of governance can be better framed as a process of building a collective capability, namely the capability of editing and managing a new kind of encyclopedia. We understand Wikipedia evolution as a learning phenomenon that gives over time rise to governance mechanisms and structures as endogenous responses to the problems and conditions that the ongoing development of Wikipedia itself has produced over the years. Finally, we put forward five empirical hypotheses to test the theoretical framework.