The study of opinion dynamics has experienced a gain in interest recently, especially with individual communication infrastructures like the Internet. Much of this research uses simulations with agents whose behavior and preferences are governed by a small set of rules. Most agent models in the literature are homophilic and/or conforming. In this work, we explore the opinion trajectory and self-sorting tendencies of a relatively new agent type: mesophilic agents, or those whose preference is for an even mix of similarity and difference of opinion with each neighbor. We characterize the behavior of these agents in topologically dynamic networks, and examine long-term behaviors of networks via common metrics as well as the reward function used to reflect agent preferences. We empirically uncover several apparent natural boundaries on things like network density and average reward score across agents that may lend themselves to a more analytical treatment moving forward.