Human Learning on the Grid will be based on the synergies between advanced software and Human agents. These synergies will be possible to the extent that conversational protocols among Agents, human and/or artificial ones, can be adapted to the ambitious goal of dynamically generating services for human learning. In the paper we highlight how conversations may procure learning both in human and in artificial Agents. The STROBE model for communicating Agents and its current evolutions shows how an artificial Agent may ”learn” dynamically (at run time) at the Data, Control and Interpreter level, in particular exemplifying the ”learning by being told” modality. The enhanced telepresence research, exemplified by Buddyspace, in parallel, puts human Agents in a rich communicative context where learning effects may occur also as a ”serendipitous” side effect of communication. The integration of the two streams of research will be the result of a workpackage within the E-LeGI EU Integrated Project, currently under negotiation.