Trust in AI is not an individual attitude but a socially negotiated, situational process embedded in team routines, with transparency, role clarity and feedback determining whether AI is used as an augmenting complement or an overlooked substitute. Poorly calibrated trust reshapes oversight, erodes informal coordination and shifts organizations toward metric-driven evaluations—so design, training and governance matter as much as algorithmic accuracy.

Main Finding

Trust in AI within project-based work is a situational, socially distributed, socio-technical process—created and maintained through ongoing boundary work between humans and machines—rather than a stable individual attitude. Trust calibration (enabled by transparency, role clarity, UX, cultural norms and system feedback) shapes collaboration patterns (e.g., delegation of oversight, erosion of informal communication) and performance outcomes (e.g., metric-driven evaluation, strategic augmentation of human expertise). The authors present an initial and a revised conceptual model linking enablers → trust-calibration practices → collaboration dynamics → project performance.

Key Points

• Theoretical framing: socio-technical systems theory (STS) and adaptive structuration theory (AST) guide the analysis of how AI integrates with team routines.
• Trust calibration is embedded in team practices and routines, not solely an individual cognitive state or attitude.
• Trust is situational and socially distributed: different team members calibrate trust differently depending on role, task and context; trust is negotiated collectively through interactions.
• Boundary work: teams continuously negotiate which inputs are treated as human versus machine, shaping responsibilities and oversight.
• Enablers of effective trust calibration: transparency/explainability, clear role definitions, good user experience, supportive cultural norms, and timely system feedback.
• Collaboration effects: typical changes include delegation of oversight to systems or specialists, changes in who communicates with whom, and erosion of informal, ad hoc communications that previously carried tacit coordination.
• Performance effects: organizations move toward metric-driven evaluation of AI outputs and often use AI to strategically augment human expertise; risks include overreliance or inappropriate metric focus.
• Contribution: advances theory by reframing trust calibration as a socio-technical mechanism linking AI design and team dynamics; provides a conceptual model for design and governance of human-centered AI in projects.

Data & Methods

• Empirical base: 40 semi-structured interviews with project professionals across multiple industries in the UK.
• Sampling: cross-industry project practitioners (roles and industries not enumerated in detail in the summary).
• Analysis: thematic qualitative analysis to derive patterns in lived experience of trust calibration, collaboration mechanisms, and perceived performance.
• Theory integration: findings interpreted through STS and AST, with theoretical extension to model trust calibration as a mediating socio-technical process.
• Nature of evidence: qualitative, interpretive—rich contextual insights but not quantitative effect sizes.

Implications for AI Economics

• Productivity and complementarities
  • Trust calibration determines the extent to which AI is used as a complement to human labor versus a substitute. Well-calibrated trust encourages complementary use (augmentation), raising effective productivity; miscalibration can lead to over/underuse and productivity losses.
• Measurement and incentive design
  • Metric-driven evaluation of AI outputs can reorient incentives and induce gaming or narrow optimization. Economists should account for shifts in measured output vs. unmeasured coordination/tacit work when evaluating AI’s productivity effects.
• Monitoring, transaction costs and governance
  • Delegation of oversight and reallocation of monitoring tasks change transaction costs within teams and firms. These shifts affect optimal organizational design, contracting, and investments in verification/audit mechanisms.
• Labor demand, skills and task allocation
  • Socially distributed trust and boundary work influence task allocation and skill requirements: demand likely increases for roles managing AI oversight, explanation, and boundary negotiation (e.g., AI integrators, translators), while routine roles may be displaced or reframed.
• Diffusion and adoption dynamics
  • Organizational norms and UX affect adoption rates. Economists modeling diffusion should incorporate social calibration processes and team-level mediators, not only individual cost–benefit calculations.
• Externalities and market outcomes
  • Erosion of informal communication and tacit coordination may impose negative externalities on team efficiency not captured in short-run metrics; conversely, improved calibration can generate positive externalities via better collective decision-making.
• Policy and regulation implications
  • Policies promoting transparency, standard feedback channels, and auditability can improve trust calibration and hence economic returns to AI investments. Subsidies for training in AI oversight roles or standards for explainability could increase welfare by reducing miscalibration costs.
• Empirical research priorities for economists
  • Quantify how trust calibration mediates AI’s effect on productivity, error rates, and team-level outcomes.
  • Estimate labor reallocation between oversight/coordination roles and task-execution roles.
  • Measure the welfare effects of metric-driven evaluation and potential gaming induced by AI integration.

Practical takeaways for economists advising firms or policymakers: evaluate AI investments not just by algorithmic performance but by anticipated effects on team routines, role clarity, monitoring needs, and measurement systems; incorporate socio-technical interventions (transparency, training, feedback loops, governance) into cost–benefit and diffusion models.