Prior work on Human-Agent Teaming (HAT) suggests AI agents integrate into teams without fundamentally altering cognitive processes. We argue that information availability creates distinct team cognition patterns in ad hoc, decision-support HATs that differ from human-only teaming. We conducted a mixed-methods study analyzing interactions across three team configurations: Human-Human, Human-Agent, and Human-Human-Agent—in resource management, finance, and health tasks, systematically varying the presence of a Human Expert and an AI Agent. Quantitative results reveal significant effects of AI agent presence on information quality, performance, and teammate preferences. Qualitative analysis using DiCoT and HAT frameworks uncovers distinct patterns in information flow, role dynamics, and turn-taking frequency, highlighting how HATs develop unique communication efficiencies and trust dynamics. Our results suggest that the variable success of HATs is due in part to team structure not being properly matched to task demands.
Yousefi et al. (Tue,) studied this question.