I present findings from a multi-year ethnographic study of automated software agents in Wikipedia. "Bots" and bot developers are a core part of the volunteer community that curates one of the world's largest and most popular information resources. The Wikipedian community relies on hundreds of independently run bots to monitor and regulate almost all aspects of the site. Bots are delegated a wide variety of organizational and administrative work, including: patrolling for spam, 'vandalism', and 'edit wars'; standardizing grammar, layout, citations, and units; updating articles using public datasets; and identifying more complicated work and distributing those tasks to humans. In my infrastructural inversion (Bowker & Star 1999), I argue Wikipedia can only appear to be governed by an economistic "wisdom of crowds" if the work delegated to bots remains invisible. These bots have long been a core way in which Wikipedians govern the 'anyone can edit' project in the absence of more formal organizational structures. Wikipedians also work out fundamental disagreements about what the encyclopedia and the community ought to look like by, in part, debating about how bots ought to be delegated governance work. For example, one of the more consistently raised (and rejected) proposals on the English Wikipedia is a bot that would make all articles conform to a single national variety of English. Lessons from Wikipedia's bots speak to many debates about how algorithmic agents are being incorporated into sites of cultural production, drawing our focus to the governance work that is delegated to automated information infrastructures.