Adaptive Human-Agent Teaming: A Review of Empirical Studies from the Process Dynamics Perspective

TL;DR

The paper addresses fragmentation in Human-Agent Teaming (HAT) research within HCI by introducing a process-dynamics lens, the $T^4$ framework, which segments the HAT lifecycle into four phases: Team Formation, Task and Role Development, Team Development, and Team Improvement. It synthesizes 133 empirical studies from 2007–2024 using a hybrid coding approach, revealing a research Fokus on Phases 2 and 3 while Phases 1 and 4 are underexplored, and highlighting core constructs such as Shared Mental Models (SMM), Mutual Theory of Mind (MToM), back-up behaviors, and dynamic delegation. The framework analyzes how task and team dynamics co-evolve, detailing agent roles (Implementer, Coordinator) and specialized roles (Advisors, Supervisors, Innovators, Learners), and discusses challenges in aligning perceptions between humans and agents. It also outlines design implications for achieving long-term adaptive HAT, including proactive agent social interaction, improved perceptual alignment, and dynamic role allocation to support self-management and resilience in real-world settings. Overall, the work provides a comprehensive blueprint for advancing long-term, adaptive HAT by bridging theory from human team dynamics with empirical HAT research, guiding future studies toward self-regulating, collaborative teams in complex environments.

Abstract

The rapid advancement of AI, including Large Language Models, has propelled autonomous agents forward, accelerating the human-agent teaming (HAT) paradigm to leverage complementary strengths. However, HAT research remains fragmented, often focusing on isolated team development phases or specific challenges like trust calibration while overlooking the real-world need for adaptability. Addressing these gaps, a process dynamics perspective is adopted to systematically review HAT using the T$^4$ framework: Team Formation, Task and Role Development, Team Development, and Team Improvement. Each phase is examined in terms of its goals, actions, and evaluation metrics, emphasizing the co-evolution of task and team dynamics. Special focus is given to the second and third phases, highlighting key factors such as team roles, shared mental model, and backup behaviors. This holistic perspective identifies future research directions for advancing long-term adaptive HAT.

Figures (7)

• Figure 1: (a) Overview of the paper search and inclusion process following the Systematic Literature Review (SLR) methodology. (b) Distribution of selected papers, with a concentration in Phase 2 and Phase 3.
• Figure 2: T$^4$ framework of HAT process dynamics. The left side illustrates the goal-action-evaluation cycle of task dynamics, which takes on different meanings at each phase of team developmental dynamics: Team Formation focuses on establishing team identity; Task and Role Development emphasizes role assignment and the task execution; Team Development is the critical phase, centering on teamwork and collaboration; Team Improvement addresses long-term sustainability and adaptability. The interplay between task dynamics and team developmental dynamics forms a process dynamics perspective spanning the entire HAT lifecycle.
• Figure 3: Capabilities and roles of agents in HAT. Existing research on agent roles in HAT follows two main threads: (a) Agent for a Team, typically in a 1vN structure, where agents act as coordinators with a focus on social capability; (b) Assistant Agent for a Task, often in 1v1 or Nv1 structures, where agents serve as implementers, emphasizing task capability. As agents gain greater autonomy, they should evolve beyond these categories to dynamically balance social and task capabilities, engaging as versatile teammates in HAT with diverse roles.
• Figure 4: Perpetual Negotiation of SMM. The left side illustrates the negotiation processes in constructing an SMM, including divergence and convergence, where divergence involves information exchange, and convergence highlights how explicit expressions and implicit cues—particularly the latter—aid in reaching temporary consensus. The right side demonstrates how continuous feedback fosters mutual learning between humans and agents, emphasizing the design mechanisms that enable agents to evolve throughout the continuous negotiation process.
• Figure 5: The perception gap between real world and perceived world. This figure illustrates the perception gap between the perceived space and three other spaces: observed space, prediction space, and explanation space. This gap arises from fundamental components of mental models—false reasoning, mismatched commitments, and inappropriate beliefs—which lead to cognitive biases, misplaced trust, and misunderstandings etc. This perception gap poses a significant challenge in constructing an SMM, while its calibration subsequently influences interaction patterns like reliance.