When AI Becomes a Teammate: Agent-Based Modeling of Triadic Human-AI Relationships

Abstract

Abstract

Recent research establishes that domain experts in triadic relationships involving human professionals, AI-based counterparts, and clients navigate trust tensions of opacity vs. performance and replacement vs. complementarity when confronted with AI systems in their professional domains (Hanelt et al., 2026). This qualitative work reveals trust building proceeds through three stages fearful exclusion, controlled opening, and opportunistic teaming driven by relational interpretation and adaptation. However, it does not examine dynamics at the team level or predict how trust emerges under varying conditions. We propose reconceptualizing the AI-based counterpart as a team member within a human-AI team, shifting the unit of analysis from the individual domain expert to the team itself, enabling investigation of collective trust emergence the process by which shared trust states co-evolve among heterogeneous agents through iterative interaction. We propose using agent-based modeling (ABM), a methodology suited to capturing emergent, nonlinear phenomena from local agent interactions (Macy & Willer, 2002). The proposed model will populate a simulated professional environment with three agent types: domain expert agents governed by trust-tension parameters, AI-counterpart agents with configurable autonomy and inscrutability, and client agents whose trust is shaped by expert endorsement and direct AI experience. Transition rules will allow agents to move between trust-building stages based on social interaction density, peer influence, self-experimentation, and relational outcomes. The model will introduce a “team trust index” as an emergent property a composite measure capturing the collective willingness of the triad to function as an integrated unit. We advance propositions for simulation testing: team trust will emerge fastest when AI experts serve as boundary spanners early on and domain experts receive low-barrier self-experimentation before client deployment, and varying autonomy-to-oversight ratios and peer network density will produce distinct trust trajectories. This work will contribute by: (1) moving the unit of analysis from individual to teams in human-AI collaboration(Anthony et al., 2023; Wang et al., 2023), (2) demonstrating ABM as an innovative IS trust methodology, and (3) offering practical guidance on sequencing social scaffolding and experimentation to accelerate collective trust.