Within the scientific field of AI, research on agent (user) modelling (AM) in dialog systems is nowadays important for several application areas (e.g., education, economics, and management) and it is related to several of its kernel subfields, such as knowledge representation, reasoning and planning. The strategic location of AM allows the construction of testbeds, such as intelligent tutoring systems (ITSs), and it supports the transfer of ideas to other larger problem domains. Our main focus has been on multiple agent societies in smart buildings, where knowledge communication plays a key role, not only among different types of agents (systems), but also between these systems and human beings (work force of an organization, maintenance officer, services engineer, security, receptionist etc.). The understanding of communication at large is crucial for designing more advanced architectures of multimodal interfaces, and it is widely accepted that user modelling is an essential component for conversations. In this paper we describe some of our experiments on modelling students and other agents keeping in mind that the results obtained so far are necessary to improve the friendliness of interactions in microsocieties.