The “smart” behaviour that characterises commercial Smart Home systems is due to the relations that users manually establish between devices and actions through automation rules. Therefore, the intelligence of these systems is entirely (or mostly) predefined by humans. To address this problem, several Artificial Intelligence techniques have been applied in this field to try to improve the smartness of Smart Home systems by enhancing features such as self-adaptation, self-evolution and self-awareness. This work in progress article presents an approach to provide smart homes with these features through a reasoning system able to deduce, define, monitor and update the automation rules that drive their behaviour, in an autonomous manner and with a knowledge management at different levels of abstraction. That is to say, the rules normally defined by humans are now created and modified automatically. The proposal is based in a commonsense knowledge model that represents how the world works and enables to infer what resources can be used to accomplish a specific objective. Automations are generated according to general, commonsense and context (topology, users, sensors, etc.) knowledge, and they are more or less sophisticated depending on the available devices. The set of behavioural rules is updated when any of these resources change. Scone is the high-performance and open-source knowledge base system where the knowledge model has been implemented. Finally, to evaluate and demonstrate the potential of the model, a set of automation rules is built for a concrete use case. Results are shown in a virtual smart home on the Home Assistant platform.