Locally Aligned, Collectively Hollowed: How LLMs Reroute Collegial Attention in Software Teams

A junior engineer asks a model instead of a senior colleague. The model answers well, the question is resolved, and the exchange between the two people does not happen. We report the human side of that coupling from 64 semi-structured interviews with designers, developers and applied AI engineers in India and the United Kingdom, conducted in three large multinational technology organisations and startups during the transition into workplace LLM use. Participants described six activities the model absorbed and, with them, six collegial acts that stopped occurring: witness, consultation, articulation, reciprocity, attribution and joint skill-building. We name the aggregate the hollowing of the collegial. The effect has no model-side observable. Each exchange is answered correctly and rated well; the loss sits in a relation the model was never party to, and no benchmark or reliance measure has a term for it. It is path dependent, because judgement that is not transmitted at the moment of the question is not transmitted later, and because the practitioners who still remember the earlier pattern will not be replaced by others who do. We give the activity-to-displacement mapping, a schematic of one rerouted attention path, and a set of collegial-cost measures for coupled-system evaluation.