What Can a Representation Support? Generative Artificial Intelligence and Professional Knowing in Practice

Abstract Generative Artificial Intelligence (GenAI) is increasingly embedded in professional university work, yet its significance cannot be inferred from individual prompts, outputs or nominal categories of use. This study examines how differently organised professional practices shape what practitioners do with GenAI and what changes, persists, or is reopened as GenAI-mediated work becomes part of continuing professional projects. Drawing on practitioner-led collaborative action research and the theory of practice architectures, we analyse a longitudinal corpus of documented GenAI-mediated interaction embedded in continuing professional projects across assessment and calibration, course development, validation and credit transfer, practicum supervision, academic supervision, and scholarly work. The findings show that professional knowing was practice-specific: what practitioners needed to distinguish and judge differed with professional project and practice architectures. Cross-practice comparison identified recurring but overlapping configurations linking professional grounds, material conditions and situated responsibility, while analytical work could be distributed to GenAI without professional mandate becoming symmetrical. Longitudinally, earlier GenAI-mediated formulations and documents could persist, be reopened, recontextualised or become consequential for later practising. Locally coherent generative work could also cease to serve a wider professional project without becoming factually false. The study therefore complicates accounts of human oversight centred on final output verification. Professional judgement was enacted throughout collaboration through negotiation, restriction, attribution, reopening, and project-level monitoring. The findings further suggest that disclosure categories or preserved chat histories do not by themselves make GenAI participation inspectable, raising implications for scholarly disclosure, student assessment, and the design of accountable AI-mediated knowledge work.