Focusing on social robots, this article argues that the form of embodiment or presence in the world of agents, whether natural or artificial, is fundamental to their vulnerability and ability to learn. My goal is to compare two different types of artificial social agents, not on the basis of whether they are more or less “social” or “intelligent”, but on that of the different ways in which they are embodied, or made present in the world. One type may be called ‘true robots’. That is, machines that are three dimensional physical objects, with three required characteristics: individuality, environmental manipulation and mobility in physical space. The other type may be defined as "analytic agents", for example ‘bots’ and ‘apps’, which in social contexts can act in the world only when embedded in complex systems that include heterogeneous technologies. These two ways of being in the world are quite different from each other, and also from the way human persons are present. This difference in ways of embodiment, which is closely related to the agents’ vulnerability and ability to learn, conditions in part the way artificial agents can interact with humans, and therefore it has major consequences for the ethics (and politics) of these technologies.