Abstract
Doing things together with generative AI involves differences with the potential for disagreement that raises analytic challenges about how to treat an LLMs outputs argumentatively. Here it is proposed that arguing with AI is most plausible when considering AI to be not simply a problem of intelligence but a puzzle of communication for argumentation. A case of extended human-AI interaction is considered to explore how an AI participates with a human through ordinary language as an interface in the interactive management of disagreement space and the uses of argument.