Stochastic parrot or not, it is less of a turkey than the colleague two desks over

This position paper contests the stabilised academic reflex that frames large language models as stochastic parrots and, by extension, as epistemically irrelevant. The paper shifts the unit of analysis from the isolated model to the sociotechnical assemblage in which the model operates. Drawing on actor-network theory, cognitive semiotics, and research on joint cognitive systems, it argues that contemporary organisations already distribute cognition across artefacts, routines, and institutional constraints, so the question is not whether a model owns human-like understanding but whether it functions as an effective mediator of knowledge and action. We integrate Paolucci’s work on humans as natural-born cyborgs and on meaning as enacted within semiotic environments to explain why human and artificial umwelten rarely intersect directly and why that mismatch is chiefly a property of biological embodiment rather than a defect of the model. We then mobilise resilience engineering and the WAx framework to describe how large language models compress complexity, accelerate conceptualisation, and support knowledge conversion across levels of work representation. The familiar catalogue of limitations is reinterpreted as a mirror of human under-specification, weak problem formulation, and institutional identity defence. The position advanced is deliberately polemical: in many modern sociotechnical systems, the most consequential bottleneck in human–model collaboration is often the human, and the emergent collaborative intelligence of the joint system can exceed that of traditional human-only teams.