In this paper, we pose and motivate a challenge, namely the need for a new science of multi-agent systems. We propose that this new science should be grounded, theoretically on a richer conception of sociality, and methodologically on the extensive use of computational modelling for real-world applications and social simulations. Here, the steps we set forth towards meeting that challenge are mainly theoretical. In this respect, we provide a new model of multi-agent systems that reflects a fully explicated conception of cognition, both at the individual and the collective level. Finally, the mechanisms and principles underpinning the model will be examined with particular emphasis on the contributions provided by contemporary organisation theory.