In this paper we critically examine the underlying assumptions in existing studies of large-scale agile software development. We use Alvesson and Sandberg’s problematization methodology and find that existing studies of large-scale agile share a number of underlying assumptions relevant to small rather than large-scale projects. Empirically, we draw on a case study of a large-scale agile project lasting nearly four years and involving more than 120 participants. Interestingly, the findings of the study contradict many of the assumptions in the literature review. For example, work across boundaries becomes at least as important as work within teams. We contribute by developing an alternative set of assumptions better suited to the characteristics of large-scale agile software development. Based on this, we re-conceptualize agile in the large, emphasizing both the complex knowledge boundaries within the project itself, as well as the interactive complexity and tight coupling with technologies and processes outside the project.