For ubiquitous computing to spread, users should be able to easily augment their cognitive environments using tags, sensors etc. However, very little is known about the psychological processes that drive such augmentation. We have developed a model of this process, and report two experiments where participants readily added 'cognitively congenial' structures to the environment for agents similar to themselves, but had difficulty when the agents were dissimilar from themselves. Our model, based on the 'simulation heuristic', explains this result, and suggests that applications like Assisted Cognition and RFID-based Robotics require interfaces that support users' mental simulation of agents and tasks.