Any agent-based model (ABM) involving agents that think or make decisions must inevitably have some model of agent cognition. Often, this cognitive model is incredibly simple, such as choosing actions at random or based on simple conditionals. In reality, agent cognition can be complex and dynamic, and for some models, this process can be worthy of its own dedicated ABM. The LevelSpace extension (Hjorth, Head and Wilensky, 2015) for NetLogo (Wilensky 1999) allows NetLogo models to open instances of other NetLogo models and interact with them. We demonstrate a method for using LevelSpace to simulate agents with complex, evolving cognitive models. We give the agents in a NetLogo predator-prey model "brains," themselves represented as independent instances of a NetLogo neural network model.