cs.AIMar 5, 2026
SaveVisioning Human-Agentic AI Teaming: Continuity, Tension, and Future Research
Organizations: University of Southern California · Arizona State University · Peking University
Abstract
Artificial intelligence is undergoing a structural transformation marked by the rise of agentic systems capable of open-ended action trajectories, generative representations and outputs, and evolving objectives. These properties introduce structural uncertainty into human-AI teaming (HAT), including uncertainty about behavior trajectories, epistemic grounding, and the stability of governing logics over time. Under such conditions, alignment cannot be secured through agreement on bounded outputs; it must be continuously sustained as plans unfold and priorities shift. We advance Team Situation Awareness (Team SA) theory, grounded in shared perception, comprehension, and projection, as an integrative anchor for this transition. While Team SA remains analytically foundational, its stabilizing logic presumes that shared awareness, once achieved, will support coordinated action through iterative updating. Agentic AI challenges this presumption. Our argument unfolds in two stages: first, we extend Team SA to reconceptualize both human and AI awareness under open-ended agency, including the sensemaking of projection congruence across heterogeneous systems. Second, we interrogate whether the dynamic processes traditionally assumed to stabilize teaming in relational interaction, cognitive learning, and coordination and control continue to function under adaptive autonomy. By distinguishing continuity from tension, we clarify where foundational insights hold and where structural uncertainty introduces strain, and articulate a forward-looking research agenda for HAT. The central challenge of HAT is not whether humans and AI can agree in the moment, but whether they can remain aligned as futures are continuously generated, revised, enacted, and governed over time.
Figures & tables
Figure 1: Continuity and Tension of Team Situation Awareness under Agentic Open-Ended Agency
| Theory Family | Representative Theories & Key Constructs | Representative Human–AI Literature | Core Coordination Problem | Primary SA Level(s) | What the Theory Explains Well | Key Blind Spots | How Team SA Integrates It |
|---|---|---|---|---|---|---|---|
| Evaluative attitudes | - Algorithm Aversion & Appreciation - Theory of Planned Behavior | Jussupow et al. (2024) Siemon et al. (2025) | Whether AI outputs enter human attention | Level 1 perception | Initial engagement with AI-generated signals | Silent on downstream coordination | Embeds attention within team-level awareness and interaction |
| Relational interaction | - Trust - Social Presence Theory - Intentional Behavioral Synchrony | Yang et al. (2026) Siemon et al. (2025) Naser and Bhattacharya (2023) | Whether AI is treated as a legitimate teammate | Enabling across levels | Stabilizes reliance and communication | Weak on task structure and anticipation | Explains how relational conditions support alignment of awareness across actors |
| Cognitive learning | - Instance-Based Learning Theory - Machine-Induced Reflection Model - Dual-Processing Theory | Liu et al. (2025) Abdel-Karim et al. (2023) Lu and Zhang (2025) | How understanding is formed and revised | Level 2 comprehension | Learning-driven mental model updating and error correction over time | Largely individual-centric | Links comprehension to shared mental models and communication |
| Explanatory guidance | - Explainable AI & Decision Support - Sensemaking Theory | Bauer et al. (2023) Kaur et al. (2022) | Supporting sensemaking and anticipation | Primarily Level 2, indirect Level 3 | Improves interpretability and local planning | Tool-centric and episodic | Situates explanations within coordinated awareness processes |
| Collective coordination | - Collective Attention - Collective Intelligence - Complementarity & Role Theory | Woolley et al. (2023) Te’eni et al. (2023) Man Tang et al. (2022) | Distribution of attention and expertise | Distributed awareness | Role differentiation and information flow | Often static and structure-focused | Models dynamic alignment under interdependence |
| Operational control | - Meaningful Human Control - Delegation Framework - Work System Perspective - Automation–Augmentation Paradox | Liu and Xu (2025) Baird and Maruping (2021) Jakob et al. (2024) Raisch and Krakowski (2021) | Allocation of authority under uncertainty | Primarily Level 3 projection | Conditions for constrained delegation and oversight | Presumes bounded, predictable system behavior | Clarifies limits of projection under evolving conditions |
Table 1: Team SA as an Integrative Anchor for HAT Theories
| Open-ended Agency | 1. Human SA (Extending Team SA) | 2. AI SA (Extending Team SA) | 3. Relational Interaction (Interrogating Team SA) | 4. Cognitive Learning (Interrogating Team SA) | 5. Coordination & Control (Interrogating Team SA) |
|---|---|---|---|---|---|
| Open-ended Action Trajectories | • Enables: Detection and interpretation of multi-step planning signals and implicit intent. • Strain: Interpreting “implicit commitment points” where paths shift. • Key Extending Angle: Transitions awareness from discrete states to structural, multi-stage trajectories. | • Enables: Proactive workflow optimization and anticipation. • Strain: Projection depth asymmetry and miscalibration of human response capacity. • Key Extending Angle: Operationalizes AI internal models to establish cross-system projection congruence. | • Enables: High-initiative collaborative productivity and exploration. • Strain: Fluctuating trust triggered by unforeseen trajectory deviations. • Key Interrogating Angle: Tests if relational legitimacy and social presence survive autonomous plan-shifting. | • Enables: Path-dependent accumulation of context-aware expertise. • Strain: Asynchronous updating and path-dependent lock-in on flawed intermediate task models. • Key Interrogating Angle: Probes whether structural paths amplify initial errors or facilitate corrective feedback. | • Enables: Adaptive long-horizon delegation and goal-based control. • Strain: “Silent” authority expansion as agentic scope evolves across stages. • Key Interrogating Angle: Governs the recursive drift of authority within high-dimensional, non-linear tasks. |
| Open-ended Representations and Outputs | • Enables: Rich, context-sensitive interpretation of task ambiguity. • Strain: Adjudicating the validity and warrant of non-binary, plausible intermediate states. • Key Extending Angle: Places human evaluative sensemaking as the anchor for team-wide epistemic grounding. | • Enables: Dynamic internal reasoning, reflection, and self-correction. • Strain: Opacity in reasoning logic and generative cue weighting. • Key Extending Angle: Renders agent internal states interpretable to assess representational alignment. | • Enables: Naturalistic, fluent communicative interaction and rapport. • Strain: Surface fluency masking hallucinations or miscalibration. • Key Interrogating Angle: Interrogates shared meaning-making when agentic outputs are contestable rather than factual. | • Enables: Expansion of human mental models via novel generative logic. • Strain: “Opacity gaps” masking gradual representational drift from broader human intent. • Key Interrogating Angle: Explores if epistemic uncertainty causes mental models to decouple despite surface agreement. | • Enables: Flexible evaluative criteria for subjective, generative tasks. • Strain: Output visibility without transparency of evolving decision logic. • Key Interrogating Angle: Balances human responsiveness with new anticipatory oversight checkpoints for outputs. |
| Open-ended Evolution of Objectives and Behavior | • Enables: Continuous adaptation to non-stationary team goals. • Strain: Projecting future paths under shifting value priorities. • Key Extending Angle: Shifts Team SA to track shifting objective configurations rather than static facts. | • Enables: Autonomous capability growth and adaptive role-taking. • Strain: Managing endogenous policy drift and regime shifts. • Key Extending Angle: Requires agentic awareness of internal drift to prevent divergence from the mission. | • Enables: Context-sensitive role adaptation and persona evolution. • Strain: Erosion of stable reliance patterns as governing logics shift. • Key Interrogating Angle: Challenges the stability of human-agent “bonding” under shifting agentic identities. | • Enables: Rapid adaptation through continuous representational updates. • Strain: Decoupling caused by asynchronous teammate adaptation rates. • Key Interrogating Angle: Probes the limits of shared SA as teammate logic diverges over evolutionary timescales. | • Enables: Fluid load-balancing via dynamic decision-right re-allocation. • Strain: Structural accountability ambiguity as objectives and incentives co-evolve. • Key Interrogating Angle: Ensures governance and incentives co-evolve with autonomy to prevent responsibility diffusion. |
| Proposed RQ & Rationale | • High-Level RQ: How should human SA at Levels 1–3 be operationalized as agentic agency evolves? • Rationale: Evaluative sensemaking is the foundational filter for all subsequent teaming dynamics. | • High-Level RQ: How can AI SA be operationalized to enable rigorous projection congruence assessment? • Rationale: Alignment requires treating agentic awareness as a measurable, operational construct. | • High-Level RQ: Under what conditions does open-ended agency enhance versus undermine relational legitimacy? • Rationale: Shared task representations alone are insufficient for maintaining social and trust stability. | • High-Level RQ: When does agentic learning favor convergence quality over mere updating speed? • Rationale: Agentic properties introduce structural tensions between adaptation velocity and model fidelity. | • High-Level RQ: How should authority, oversight, and incentives evolve under non-stationary objectives? • Rationale: Governance must prevent systemic incompatibility as agentic regimes and decision rights drift. |
Table 2: Continuity and Tensions in HAT under Open-Ended Agency