Researchers claim that pedagogical agents engender opportunities for social learning in digital environments. Prior literature, however, has not thoroughly examined the discourse between agents and learners. To address this gap, we analyzed a data corpus of interactions between agents and learners using open coding methods. Analysis revealed that: (1) conversations between learners and agents included sporadic on-task interactions with limited follow-up; (2) conversations were often playful and lighthearted; (3) learners positioned agents in multiple instructional/social roles; (4) learners utilized numerous strategies for understanding agent responses; (5) learners were interested in agents' relationship status and love interests; and (6) learners asked personal questions to the agent but did not reciprocate to requests to talk about themselves.