Human-AI teams succeed when systems enable coordination, clear expectations and adjustable autonomy rather than merely dividing tasks. Yet the literature is fragmented and skewed toward lab demonstrations, calling for more longitudinal, real-world studies to judge sustained workplace impact.

From testbeds to high-stakes work: a review of Human-AI teaming domains and teaming factors

Shaida Kargarnovin, C. Hernandez, D. Reiners, C. Cruz-Neira, G. Bochenek, Waldemar Karwowski · Fetched June 22, 2026 · Frontiers in Robotics and AI

semantic_scholar review_meta medium evidence 7/10 relevance Full text usable extracted full text DOI Source PDF

A PRISMA-guided review of 104 empirical studies finds human-AI teaming yields best outcomes when systems support coordination, calibrate transparency, and allow adjustable autonomy, but the evidence base is fragmented and dominated by controlled studies rather than longitudinal, in-context evaluations.

Introduction Human-AI teaming is increasingly being studied in applied and high-stakes settings, yet the evidence remains dispersed across domains, constructs, and research traditions. This fragmentation also limits efforts to connect broader human-AI findings to human-robot teaming (HRT), where embodied systems make issues such as coordination, autonomy management, communication, and safety more immediate in real-world interaction. Methods To provide a clearer picture of the field, we conducted a PRISMA-guided systematic review with bibliometric analysis of 104 peer-reviewed empirical studies published between 2015 and 2025 and identified through Engineering Village, IEEE Xplore, PubMed, ScienceDirect, and Web of Science. Results The review maps where human-AI teaming has been evaluated and what teaming aspects are most frequently examined. Cross-domain and interdisciplinary studies were the largest category, representing broad workplace or team-based investigations not tied to a single industry and instead focused on general collaboration issues such as communication, teamwork, coordination, and coworker interaction. Gaming and entertainment, aviation, military and defense operations, emergency response and public safety, and healthcare also represented substantial portions of the literature. Across studies, performance was the most frequently examined aspect, followed by trust, explainability and transparency, decision-making, and team processes. Bibliometric patterns suggest a shift since 2020 from foundational demonstrations in controlled settings toward applied, higher-stakes contexts where trust dynamics, communication, and ethical accountability more directly shape adoption and sustained performance. Discussion Evidence points to a practical conclusion that human-AI teaming works best when the interaction supports coordination, allowing users to form accurate expectations of the AI, adjust autonomy and delegation across task phases, and use transparency cues that calibrate reliance without adding burden. For HRT, these findings reinforce the importance of shared control, mixed-initiative interaction, and designs that help humans and robots coordinate action over time rather than simply divide functions. We conclude by outlining implications for designing and evaluating human-AI teams as socio-technical systems and for prioritizing longitudinal and in-context studies that capture how teaming evolves over time.

Summary

Title: From testbeds to high-stakes work: a review of Human-AI teaming domains and teaming factors Citation: Kargarnovin S. et al., Frontiers in Robotics and AI (2026). DOI: 10.3389/frobt.2026.1733942

Main Finding

Human-AI teaming research (104 empirical studies, 2015–Aug 2025) has moved from controlled testbeds toward applied, higher-stakes domains since ~2020. Across domains, studies most often evaluate performance, trust, explainability/transparency, decision-making, and team processes. For embodied settings (human-robot teaming, HRT), coordination, shared control, mixed-initiative interaction, autonomy management, and safety emerge as central design requirements to enable reliable, sustained teaming.

Key Points

Evidence base and scope
- 104 peer‑reviewed empirical human-subject studies (2015–Aug 2025) identified via PRISMA-guided search across Engineering Village, IEEE Xplore, PubMed, ScienceDirect, Web of Science plus backward citation searches.
- Inclusion limited to English, peer-reviewed, human experiments examining human-AI teaming.
Domain distribution (primary domain assigned per study)
- Cross-domain / interdisciplinary workplace studies: 22%
- Gaming & entertainment: 19%
- Aviation, military & defense: 12%
- Emergency response & public safety: 11%
- Healthcare: 9%
- Remaining studies cover transportation, engineering design, and other areas.
Most-studied teaming aspects (descending frequency)
- Performance (most frequent)
- Trust
- Explainability / transparency
- Decision-making
- Team processes (communication, coordination, role allocation)
Temporal and bibliometric trends
- Pre-2020: emphasis on foundational demos and controlled lab testbeds.
- Post-2020: shift toward applied, in-context, higher-stakes work where trust dynamics, communication, autonomy, ethics, and accountability drive outcomes.
HRT-specific insights
- Embodiment raises coordination, spatial/temporal alignment, and physical-safety concerns not present in purely disembodied AI.
- Effective HRT requires shared mental models, anticipation/prediction capabilities, mixed-initiative control, and mechanisms for autonomy adjustment across task phases.
Practical design takeaways
- Systems that support coordination (not just task division) and enable humans to form accurate expectations perform better.
- Transparency cues should calibrate reliance without imposing cognitive burden.
- Autonomy management (adjustable autonomy, delegation) across phases is important for sustained performance.
Study limitations reported by authors
- No formal cross-study quality scoring due to heterogeneity of methods and settings.
- English-only, peer-reviewed empirical studies; timeframe limited to Jan 2015–Aug 2025.

Data & Methods

Review type: PRISMA-guided systematic review with complementary bibliometric analysis.
Data sources: Engineering Village, IEEE Xplore, PubMed, ScienceDirect, Web of Science; backward citation search.
Search window: January 2015 – August 2025.
Selection criteria: empirical human-subject experiments investigating human-AI teaming; peer-reviewed; English.
Records: 1,084 initial records → 957 after duplicates → 104 included studies after screening and eligibility checks.
Synthesis: studies coded by primary application domain and focal teaming elements; bibliometric trends analyzed to reveal temporal shifts in research emphasis.
Quality appraisal: no single numeric quality score applied due to methodological heterogeneity; rigor maintained by strict inclusion criteria.

Implications for AI Economics

Labor complementarities and productivity
- Findings support that AI as teammate can enhance productivity when tasks permit coordination and complementary roles (humans: judgment/creativity; AI: pattern recognition, scale). Economic models should treat AI as a collaborative input that changes marginal productivity depending on coordination frictions and team-process design.
Adoption frictions & demand-side constraints
- Trust, explainability, and calibrated autonomy matter for adoption—especially in high-stakes industries. Firms’ adoption decisions will depend not only on expected performance gains but on perceived reliability, liability exposure, and cognitive costs to workers.
Human capital & task reallocation
- Effective teaming requires human skills (coordination, supervision, mental-modeling of AI). Investments in training and re-skilling are necessary for realizing productivity gains; models of labor transitions should incorporate costs and time lags of such investments.
Value of explainability & transparency
- Explainability and transparency have measurable economic value by reducing misuse/over-reliance, improving trust calibration, and lowering monitoring costs. This implies a nontrivial return to investing in explainability features and standards—especially where liability or regulatory scrutiny exists.
Risk, liability, and insurance markets
- Higher-stakes deployments (healthcare, aviation, emergency response, defense) amplify the economic importance of safety, liability allocation, and insurance pricing. Empirical evidence on how teaming designs affect error modes can inform contract design, regulation, and insurers’ risk assessments.
Organizational design & procurement
- Successful teams require socio-technical integration (interfaces, autonomy management). Procurement decisions (e.g., in government or large firms) should factor lifecycle costs: training, human–AI coordination overhead, and longitudinal performance rather than single-run benchmarks.
Market structure & competition
- Firms that internalize coordination design—investing in interfaces, mixed-initiative control, and longitudinal evaluation—may capture more value from AI teammates, suggesting first-mover advantages in sectors with complex teaming needs. Standardization around transparency and autonomy interfaces could lower entry barriers.
Evaluation metrics & ROI measurement
- Because many studies emphasize in-context and longitudinal dynamics, economic evaluation should move beyond short-term task performance metrics to measures of sustained output, error rates over time, turnover, litigation risk, and learning curves.
Policy and regulation considerations
- Evidence supports targeted regulation that mandates transparency and human-control affordances in high-stakes AI deployments. Economic analyses should consider compliance costs, potential liability transfers, and incentives for safer design.
Research gaps with economic relevance
- Few longitudinal field studies quantify productivity, wages, or labor reallocation due to human-AI teaming in real workplaces. Economists should prioritize causal field evidence on wages, employment composition, training costs, and firm-level productivity responses to teaming interventions.

Suggested next steps for AI economists - Design randomized or quasi-experimental field studies that measure firm-level productivity, employment outcomes, and worker welfare when AI teammates are introduced with differing transparency and autonomy regimes. - Quantify returns to explainability and training investments in real deployments; model trade-offs between automation levels and supervision costs. - Integrate coordination frictions and team-process adjustment costs into models of AI adoption and diffusion across industries, with heterogeneous worker skill distributions and regulatory environments.

If you’d like, I can convert these implications into a short research agenda (specific hypotheses, data sources, and empirical designs) tailored for empirical economists.

Assessment

Paper Typereview_meta Evidence Strengthmedium — The paper synthesizes 104 empirical studies across domains and provides consistent patterns (coordination, calibrated transparency, adjustable autonomy) but the underlying evidence is heterogeneous, often from controlled lab experiments rather than randomized field trials or longitudinal in-context studies, limiting causal inference and external validity. Methods Rigorhigh — PRISMA-guided systematic review combined with bibliometric analysis across five major databases (Engineering Village, IEEE Xplore, PubMed, ScienceDirect, Web of Science) and a clearly defined 2015–2025 window demonstrates rigorous, reproducible literature search and synthesis procedures, though conclusions remain constrained by the quality and design of included studies. SampleA systematic sample of 104 peer-reviewed empirical studies published 2015–2025 identified via Engineering Village, IEEE Xplore, PubMed, ScienceDirect, and Web of Science; studies span cross-domain workplace/team investigations, gaming/entertainment, aviation, military/defense, emergency response/public safety, and healthcare, and examine outcomes such as performance, trust, explainability, decision-making, and team processes. Themeshuman_ai_collab org_design productivity adoption GeneralizabilityMany included studies are controlled lab or simulation experiments, reducing ecological validity for real-world workplaces, Few longitudinal or in-situ studies capture how teaming evolves over time, Substantial heterogeneity across domains and tasks complicates generalizing findings to specific industries, Potential publication and English-language/database selection bias (peer-reviewed outlets searched), Measures and operationalizations of teaming constructs vary across studies, limiting comparability

Claims (8)

Claim	Direction	Outcome	Confidence & Evidence	Details
We conducted a PRISMA-guided systematic review with bibliometric analysis of 104 peer-reviewed empirical studies published between 2015 and 2025 and identified through Engineering Village, IEEE Xplore, PubMed, ScienceDirect, and Web of Science. Other	null_result	number_of_studies_reviewed	Reading fidelity high Study strength high	n=104 0.4
Cross-domain and interdisciplinary studies were the largest category, representing broad workplace or team-based investigations not tied to a single industry and instead focused on general collaboration issues such as communication, teamwork, coordination, and coworker interaction. Other	null_result	study_domain_prevalence	Reading fidelity high Study strength medium	n=104 0.24
Gaming and entertainment, aviation, military and defense operations, emergency response and public safety, and healthcare also represented substantial portions of the literature. Other	null_result	study_domain_representation_by_industry	Reading fidelity high Study strength medium	n=104 0.24
Across studies, performance was the most frequently examined aspect, followed by trust, explainability and transparency, decision-making, and team processes. Team Performance	null_result	performance (and ranked prevalence of constructs like trust, explainability, decision-making, team processes)	Reading fidelity high Study strength medium	n=104 0.24
Bibliometric patterns suggest a shift since 2020 from foundational demonstrations in controlled settings toward applied, higher-stakes contexts where trust dynamics, communication, and ethical accountability more directly shape adoption and sustained performance. Adoption Rate	null_result	temporal_shift_in_research_contexts (prevalence of applied/higher-stakes contexts post-2020)	Reading fidelity high Study strength medium	n=104 0.24
Evidence points to a practical conclusion that human-AI teaming works best when the interaction supports coordination, allowing users to form accurate expectations of the AI, adjust autonomy and delegation across task phases, and use transparency cues that calibrate reliance without adding burden. Team Performance	positive	teaming_effectiveness / performance as a function of coordination, expectation calibration, adjustable autonomy, and transparency cues	Reading fidelity medium Study strength medium	n=104 0.14
For human-robot teaming (HRT), these findings reinforce the importance of shared control, mixed-initiative interaction, and designs that help humans and robots coordinate action over time rather than simply divide functions. Team Performance	positive	coordination_and_control_design_effects_on_HRT_performance	Reading fidelity medium Study strength medium	n=104 0.14
We conclude by outlining implications for designing and evaluating human-AI teams as socio-technical systems and for prioritizing longitudinal and in-context studies that capture how teaming evolves over time. Research Productivity	null_result	research_design_priorities (longitudinal and in-context evaluation)	Reading fidelity high Study strength speculative	n=104 0.04