Observation is a fundamental interaction pattern in today's computer-based systems. Adopting observation as the main modelling criterion, computer-based systems can be represented as composed by three class of entities: observers, observables (or sources), and coordinators, that is, the entities managing the observer/source interaction.

Also, agents and agent societies are fundamental abstractions in modelling today's complex systems. When exploiting observation in the context of agent-based systems, the most natural interpretation for agents is to see them as either observers or coordinators. However, their situatedness and autonomy, their peculiar perception and representation of the environment, and their typical ability to infer new knowledge - in short, their individual viewpoint over the world -, make agents suitable for an interpretation as observable sources.

Accordingly, this paper discusses the implications of using observation to model agent systems, and focuses on the interpretation of agents as observables. A formal framework is developed where multiagent systems are modelled as the composition of agents interacting by observing each other and by mutually affecting their observable behaviour.