Multiple devices, both hardware and software, may come and go at any time in a given room. Software controlling the behaviour of these devices must be able to adapt to encompass new devices or the removal of existing devices. This paper presents a model for curious, supervised learning agents that address the issue of adaptability at a behavioural level in an intelligent room. Curious, supervised learning agents comprise a curiosity module and a supervised learning algorithm. The curiosity module identifies interesting devices on which to focus the agent’s learning. The supervised learning component realises behaviours by observing, modelling and mimicking human actions. Our framework is demonstrated in a virtual meeting room in Second Life. We show that the curious learning agent can adapt its behaviour to identify new learning goals in response to new devices and activities.