In this paper, we present a novel approach for tuning power modes of wireless 802.11 interfaces. We use K-means and simple correlation techniques to analyze user's interaction with applications based on mouse clicks. This provides valuable contextual hints that are used to anticipate future network access patterns and intent of users. Based on those hints, we adapt the power mode of the wireless network interface to optimize both energy usage and bandwidth usage. Evaluation results (based on real data gathered from interaction with a desktop) show significant improvements over earlier power management schemes.