Social networks present distinctive features when compared with other types of networks, particularly the presence ofcommunities, which are subsets of nodes much more densely connected among themselves, than with the rest of the net-work. In this work, we propose an explanation for this pattern based on the following: groups may be the communitysolution of hunter-gatherer societies to the survival problem posed by the uncertainty of food. We propose a multi-agentmodel inspired by a food-sharing dynamic, which combines and formalizes two main notions discussed by some anthropo-logical literature: the reciprocity in the exchanges of food, plus the care for the general welfare of agents. Our preliminaryresults show that near-to-optimal food-sharing networks exhibit highly-connected groups around special agents that wecall hunters, those who inject food into the system. We show the robustness of these results by computer simulations andalso by analytical arguments for these simulations.