What blocks effective human-AI teams from performing well?

"Unraveling Human-AI Teaming: A Review and Outlook" argues that human-AI collaboration remains "ad hoc, often resulting in underwhelming performance" because the information-systems literature still "mainly treats AI as a tool rather than a nearly full-fledged teammate." The authors name two research gaps: first, "a significant challenge in designing AI agents that align with human values and objectives," which they trace partly to the fact that "humans themselves often struggle to explicitly articulate their goals and objectives"; second, "the limited integration of AI's full capabilities into collaborative workflows," since "many still view AI agents as static tools rather than adaptive, knowledge-enhancing partners," reinforced by "behavioral resistance and inertia" among people reluctant to treat AI as a genuine team member. To address both gaps the paper proposes a research outlook built around four aspects of human-AI teaming: "formulation, coordination, maintenance, and training."

The theoretical scaffold is Team Situation Awareness (SA) theory (Endsley 1995), which the paper extends rather than discards. Individual SA already has three levels — perception, comprehension, and projection — and the classic model adds "shared SA," defined as "the degree to which team members possess the same SA on shared SA requirements," resting on a "shared mental model." The paper's extension is to add "role specification and role fluidity" to this model so it covers AI teammates: role specification ensures "all team members, including AI, clearly understand their specific tasks," while role fluidity is the "adaptability, open communication, and shared responsibilities" needed to "respond effectively to dynamic and uncertain environments." The authors treat trust as the mechanism that makes this work in practice, citing Salas et al. (2017) that trust "is essential to enhancing information sharing and reducing communication hesitancy," so that trust growing as AI "contributions become more accurate and contextually relevant" should enable "smoother workflows and more confident task delegation."

This sits as a theoretical companion to two nearby position papers rather than a test of either. Where Does machine agency exist on a spectrum rather than binary? locates AI's role on a spectrum from passive tool to cooperative partner at the level of individual human-machine interactions, this paper works at the team level, asking what shared awareness and role structure a team needs once AI sits somewhere past "passive tool" on that spectrum. It also echoes the claim in Should AI systems stay collaborative rather than fully autonomous? that human involvement remains structurally necessary — but where that paper grounds the claim in autonomous agents' reliability and transparency failures, this paper grounds the same conclusion in Team SA theory's requirement for shared mental models and trust-building, a different argument for a similar position. The four-aspect agenda (formulation, coordination, maintenance, training) also offers a theory-first complement to the empirically derived role typologies in Do humans and AI divide labor differently in knowledge work? and How do AI novelty and scope shape employee relationships?, both of which describe role divisions that could be read as instances of the "role specification and role fluidity" this paper proposes in the abstract.

The excerpt is explicitly "a foundational agenda for future research," not a tested model — the role-specification and role-fluidity extensions to Team SA are proposed, not validated against any human-AI team, and the paper names the value-alignment gap without resolving it. The Discussion section also flags, citing Dell'Acqua et al. (2023) and Fügener et al. (2021), that "over-reliance on AI may suppress human creativity, learning, and critical thinking" as trust grows — a risk the paper raises as a tension to manage rather than something it measures. The implication is that the four-aspect research outlook is best read as a set of hypotheses for future human-AI teaming studies to test, not evidence that current AI deployments already achieve shared situation awareness with their human teammates.