In recent years a number of techniques have been studied for augmenting the ease of use of 3D worlds: methodologies for adapting both navigation and content allow a user to interact with a customized 3D world; adaptable navigation paradigms offer parallel modalities for different classes of users. In both cases the goal is to reduce the cognitive load needed for interaction.This work focuses on interaction adaptivity, trying to anticipate the user behaviors by monitoring their interaction patterns. Identification of recurring patterns of interaction is used for shortening users interaction when they approach objects belonging to the same class of those ones previously accessed.An agent based approach is used for monitoring the user activity and for proactively adapting interaction: a pair of agents, called numen (bound to the user) and genius loci (associated with the interaction areas called interaction loci) observe and mediate the users' interaction and adapt the 3D world in order to match their needs.The users are always made aware by the system about the changes in the 3D world and they are always in control. They can deactivate adaptiveness for a while, letting the system only to monitor their actions without applying automatically interaction shortcuts. An implementation architecture and a demonstration prototype are presented. A case study related to virtual fairs is examined, discussing a number of adaptive interaction examples.