A core problem of social constructivist theories of knowledge is the lack of theoretical clarity about the role of knowledge constraints and how they are overcome in practice, by what type of social agency. Knowledge constraints are both special - rules constituting and/or regulating such distinct intellectual fields as science, art and technology - and general - laws working across the nature/culture divide. In order to sort out this complexity of knowledge constraints in science, we need to start by recognizing the existence of knowledge constraints in the first place. Advance of knowledge is not contingent, as claimed by the Strong Program inaugurated by David Bloor. The latter reduces the problem of knowledge constraints to reputational work. But reputational work is what all humans engage in; this is one reason why networking is so crucial for the human species. Moreover, it requires the special form of communicative competence, talk, which only humans possess. It is also because humans can talk that they have managed to invent the coherent argument, the core of the special constraint that constitutes science, giving it a law-like character. Arguments are both competitive and cooperative at the same time, reputational work contains both elements. This strongly social element of science is missing in Latour's concept of the social. Latour reduces the social (both networking and agency) to naked competition. But the model for this thinking is not sociology, it is literature. As a social construction it possibly reflects the psychological effect of the French system of elite education.