The position of an access point (AP) in a WiFi network has considerable influence on the performance of the network. In this work, we consider the problem of a WiFi AP self-positioning itself adaptively based on the network conditions to deliver improved network performance. Through extensive experimental evaluation, we show that there are indeed significant performance benefits to be attained by allowing the AP to move intelligently. We also rely on theoretical analysis, simulations, and experimental studies to show that the AP optimal location search problem can be split into two parts: a macro-search problem to minimize average path loss between AP and clients, and a micro-search problem to tackle real-time multipath fading effects. We then present Hermes, a self-positioning WiFi AP system that relies on a suite of algorithms to compute and then move to an optimal location within the network. Using a prototype implementation, we show that Hermes can perform up to 117% better than WiFi with no AP mobility, and up to 73% better than related work that allows for AP mobility.