The problem of interest is how to dynamically allocate wireless access services in a competitive market which implements a take-it-or-leave-it allocation mechanism. In this paper we focus on the subproblem of preference elicitation, given a mechanism. The user, due to a number of cognitive and technical reasons, is assumed to be initially uninformed over their preferences in the wireless domain. The solution we have developed is a closed-loop user-agent system that assists the user in application, task and context dependent service provisioning by adaptively and interactively learning to select the best wireless data service. The agent learns an incrementally revealed user preference model given explicit or implicit feedback on its decisions by the user. We model this closed-loop system as a Markov Decision Process, where the agent actions are rewarded by the user, and show how a reinforcement learning algorithm can be used to learn a model of the user?s preferences on-line in the given allocation mechanism. We evaluate the performance and value of the agent in a series of preliminary empirical user studies.