Coordination, Adaptation, and Learning in Human, AI, and Hybrid Teams: Towards a New Research Frontier
Abstract
Autonomous teams composed of generative AI agents, specifically large language models (LLMs), now perform interdependent work alongside human peers. We examine what this implies for process theories of teams, taking McGrath and Argote’s ( 2002 ) conceptualization of coordination, adaptation, and learning as our foundation. Their framework already admits software as a tool, knowledge held outside members, processes of differing cadence, and learning retained in non-human repositories. We retain it and pose six questions regarding substrate, coordination, temporal structure, goal alignment, adaptation, and learning. Three public cases (Cursor, MetaGPT, and TheBotCompany), chosen for differences in who may alter a team’s coordination structure and when, illustrate these questions without testing them. We theorize about three constructs: an authority record for each relation in a mixed coordination network; temporal pluralism, coupling execution pace, review latency, and the point after which downstream work builds on a change, so the abstract matches the new definition; and cross-substrate translation, the transfer of retained change across repositories of different kinds. We distinguish six operations commonly termed AI learning, from parameter updating to architectural change. Three propositions are restricted to cooperative, orchestrated teams of fixed-weight LLM agents under one mandate. We conclude with limitations and guidance.