Along evolution, increasingly complex cognitive functions have been attributed to an increasingly complex brain architecture. Nevertheless, the brain remains anchored on an organization dedicated to survival. We believe that keeping this principle in mind is an excellent way to better decipher cerebral mechanisms and corresponding cognitive functions. Accordingly, we describe here the main characteristics and constraints of an intelligent agent learning to survive in an intelligent environment, in terms of information flows and learning principles. On this basis, we propose a framework of description for the architecture of the brain of mammals, organized around four fundamental questions to be answered. These questions define the identity of the goal (what ?) and the motivation to choose it (why ?), its location (where ?) and the way to get it (how ?). Then we explain how the main requirements of respondent and operant conditioning can be addressed within this architecture and how it is also compatible with the elaboration of more complex cognitive mechanisms. This can be seen as the validation of this framework to explore how cognitive functions might emerge from cerebral circuits and how they have been made more complex along evolution. It also proposes a systemic view of the brain, useful to develop the cogni-tive architecture of an intelligent agent exploring autonomously its environment and to propose to machine learning innovative algorithms.