We develop a game-theoretic model to analyze optimal workplace arrangements in AI-enhanced teams where knowledge sharing is subject to location-dependent costs. Extending principal-agent theory to incorporate remote collaboration frictions, our model shows how return-to-office (RTO) policies affect incentives for employee effort and AI knowledge transfer. We identify conditions that ensure efficient in-person and remote work arrangements. Comparative statics show that higher AI adoption paradoxically reduces tolerance for remote work when it increases the frequency of costly knowledge-sharing events. Conversely, AI...