The Collaboration Gap in Human–AI Work

LLMs are increasingly presented as collaborators inprogramming, design, writing,and analysis. Yet the practical experience of working withthem often falls short of this promise. In many settings,users must diagnose misunderstandings, reconstruct missingassumptions, and repeatedly repair misaligned responses.This poster introduces a conceptual framework forunderstanding why such collaboration remains fragile.Drawing on a constructivist grounded-theory analysis of 16interviews with designers, developers, and applied AIpractitioners working on LLM-enabled systems, and informedby literature on human–AI collaboration, we argue thatstable collaboration depends not only on model capabilitybut on the interaction’s grounding conditions. Wedistinguish three recurrent structures of human–AI work:one-shot assistance, weak collaboration with asymmetricrepair, and grounded collaboration. We propose thatcollaboration breaks down when the appearance ofpartnership outpaces the grounding capacity of theinteraction and contribute a framework for discussinggrounding, repair, and interaction structure in LLM-enabledwork.