Autonomy is regarded as a crucial notion in multi-agent systems and several researchers have tried to identify what are the agent's parts that give it an autonomous character. In this paper, we take a different approach. If we assume that agents are autonomous (and this is a quite reasonable assumption in many practical situations, such as e-commerce), the more interesting question is: how to cope with the autonomy of agents? What are the effects on the way agents have to coordinate their behavior with other agents, and on the agent design process? And what are the effects of that (secondary effects) on the architecture of agents and agent societies. We address these questions by working out the concept of „collaboration autonomy“, and by describing an agent society model that respects this kind of autonomy.