AI can be the object of a study, the instrument of a study, or both at once, and which of the three is in play changes the design, the threats to validity, the ethics, and what has to be reported. This session does not hand over the distinction as a definition: participants apply it to eight real and constructed studies, and discover that the studies that look most obviously like one thing are often also the other. We then read published research on AI as a method, through one lens (unit of analysis, what is manipulated, what could make the result false), and attack two studies in small groups to see how research on AI and research with AI break in different ways. The afternoon closes on the apparatus around a model rather than the model itself, and on each group’s own research question, which becomes the starting point of the final project.