This paper studies the impact of diversity in cognitive ability among members of a team on their performance. We conduct a large field experiment in which teams start up and manage real companies under identical circumstances. Exogenous variation in - otherwise random - team composition is imposed by assigning individuals to teams based on their measured cognitive abilities. The setting is one of business management practices in the longer run where tasks are diverse and involve complex decision-making. We propose a model in which greater ability dispersion generates greater knowledge for a team, but also increases the costs of monitoring necessitated by moral hazard. Consistent with the predictions of our model, we find that team performance as measured in terms of sales, profits and profits per share first increases, and then decreases, with ability dispersion. Teams with a moderate degree of ability dispersion also experience fewer dismissals due to fewer shirking members in those teams.