I defend a form of pragmatic encroachment in epistemology. In particular, I argue that what someone knows is sensitive to which questions they are answering. The natural way to spell out this kind of approach has been recently shown to have a serious problem. It allows agents to know conjunctions when they don't know either conjunct. Worse still, the obvious way out of the problem yields an implausible form of scepticism. I will argue that the problem can be elided if we are careful about just what question we take agents to be answering when they face a practical choice.