Why do governments and industry contract voluntary agreements (VA) for the improvement of the environment? The number of agreements in this field is constantly increasing and an important one was recently signed in France on carwaste processing. This agreement aims to trigger the emergence of recycling as a new value for end-of-life vehicles (ELV) and for car design. It implies the expected commitment of all professionals of an industry in the concurrent development of new technologies, new markets and new firms without any state subsidies. The theoretical interpretation of this process challenges the classical approaches of regulation, particularly the regulatory capture theory and its agency-theory extensions. An explanation of the genesis of such agreements, their impacts and robustness requires us to root agency approaches in three theoretical and dynamic perspectives :

H1 : an “operational” theory of legal regulation ;

H2 : market transactions seen as “reciprocal prescriptions”;

H3 : firm’s strategy formulation as an organizational learning process.

Within this framework we can understand why, in specific contexts of uncertainty and complexity, voluntary agreements are both a result and an instrument of the regulatory process. They avoid the pitfalls of a law. They may also stimulate co-ordinated learning and exploratory actions in a competitive context even when there are no strong economic incentives or assessed technological solutions at the outset to bring about a collective improvement. The same theoretical perspectives can highlight the drawbacks that threaten such agreements and indicate a dynamic pattern of increasing interventionism. A general regulation strategy can thus be defined in three phases: 1) establishment of a VA; 2) if necessary, reinforcement through legal standards and policy monitoring; 3) if – and only if – environmental goals are not attained, taxes and regulation might be considered. The latter instruments are applied when enough knowledge is available to limit perverse effects. This strategy is profoundly meaningful: public intervention requires knowledge, but also guides the production of knowledge. Hence, any intervention strategy has to be based on a specific and explicit pattern of collective learning. The carwaste case, studied by the authors over the past four years with Renault, is used to illustrate these different points. A new model of state intervention is derived from this case and its validity is explored through a comparative study with other cases.