Since a couple of years, new approaches are proposed to solve combinatorial optimization problems: multi-agent systems. In this paper, we propose a new model, EC4MAS, to build self-organizing multi-agent systems with more endogenous control. We start presenting a representative set of solving methods and we highlight what are the key elements of these solving processes and how they are used to construct a new representation of the problem to solve it. Generally, this representation is based on the characteristics of the method implemented but the construction of this representation could happen in the system without so much external intervention. This has been illustrated by some work in psychology that we present. Based on this observation, we propose and illustrate in this work a self-organizing multi-agent approach that tries to construct itself this representation, in an endogenous way. It is organized into a social organization of the different local solving behaviors and a spatial organization that represents the different structural/topological characteristics of the problem. The objective of the system is thus to find a good coupling between these two organizations to get the best possible representation.