Research libraries continue to reinvent themselves in the face of increasing demand from users for digitized texts. As physical books move from stacks to deep storage, many researchers lament the reduction in the serendipitous discovery that was provided by browsing the stacks. We believe, however, that digitization offers even greater opportunities for guided serendipity. Developments in machine learning and computing at scale allow content-based models of library collections to be made accessible to patrons. In this talk, we will present a vision for the future of library browsing using the Topic Explorer â€°Ã›ÃHypershelfâ€°Ã›Â that we have developed for digital collections. It allows users to jump into the collection and browse nearby volumes, rearranging them at will according to topics extracted computationally from the full texts. We will demonstrate the Hypershelf in action, and discuss how it might be integrated with physically-shelved books. This vision enhances rather than supplants the traditional librarians' function of guiding patrons to the best starting points for their research needs.