As a theoretical computer scientist, one comes across the most curious tasks when looking at other disciplines: In the humanities, for instance, one can find poems written on palm leaves that are interspersed with comments from over the centuries. And if a humanities student does not know the poem by chance, they have a hard time identifying it in the text. Solving this spe-cific challenge has lead to an agent-like system with a corpus of annotated poems at its centre. Its task is to partition an unknown document into poem and comments. It uses a light-weight combination of well-established methods based on vector spaces as well as probabilistic graphical models to solve the task. Such information retrieval systems always serve a user's, i.e., a human's, purpose. As such, it represents a very specific instance of human-aware AI, which deals with inference or planning problems in collaboration with or on behalf of a human and aims at interpretable or explainable behaviour. This talk journeys from this specific digital humanities task to human-aware AI. It thereby demonstrates how keeping a human in the for-malism helps keep researchers on their toes.