The concept of autonomous artificial agents has become a pervasive feature in computing literature. The suggestion that these artificial agents will move increasingly closer to humans in terms of their autonomy has reignited debates about the extent to which computers can or should be considered autonomous moral agents. This article takes a closer look at the concept of autonomy and proposes to conceive of autonomy as a context-dependent notion that is instrumental in understanding, describing and organizing the world. Based on the analysis of two distinct conceptions of autonomy, the argument is made that the limits to the autonomy of artificial agents are multiple and flexible dependent on the conceptual frameworks and social contexts in which the concept acquires meaning. A levelling of humans and technologies in terms of their autonomy is therefore not an inevitable consequence of the development of increasingly intelligent autonomous technologies, but a result of normative choices.