Recent research results in Theory of Actions, Database Updating, Belief Revision and Abduction employ stronger, more reasonable and constructive forms of Nonmonotonic Reasoning. However these results do not account for a plethora of basic features of an intelligent system or agent. We propose the study of these formalisms and their extensions by means of more conceptually satisfactory structures, such as metric spaces, and syntactical representations of them. We believe that this study will bring us closer to a description of an intelligent system by taking into account its global preferences. Our approach will b resource-conscious by studying its complexity and by logically characterising it with simple rules. Links with applications to Artificial Intelligence and Knowledge Engineering, such as Planning and Diagnosis, will be investigated.