Psychological safety in AI-augmented teams: a longitudinal cohort study
How psychological safety in teams changes when AI teammates are introduced, and how safety climate impacts team performance and adoption outcomes.
Authors
S. Andersen, HERMES, 20 contributors
Published
2030
Citations
88
Overview
Psychological safety—the belief that you can take interpersonal risks without fear of negative consequences—is foundational to team performance. When AI is introduced as a team member, does safety increase (less risk of human error) or decrease (fear of AI, job security concerns)? This 18-month study tracked 52 teams.
Methodology
Longitudinal cohort design tracking 52 teams (780 individuals) over 18 months. Monthly psychological safety surveys using validated scales (Edmondson PSS). Performance tracking through productivity metrics, error rates, and learning velocity. Qualitative interviews with 120 team members at 3, 6, 12, and 18-month marks.
Key Findings
Psychological safety decreases by average of 28% in first 60 days after AI teammate introduction, then recovers by month 6 in well-managed transitions but remains 12% below baseline in poorly-managed ones. The initial safety decline reflects uncertainty and role ambiguity, not fundamental fears about AI.
Teams with explicit role clarity (defining what humans and AI do) maintain psychological safety 41% better than teams without explicit role definitions. Similarly, teams with regular feedback loops (weekly or more) about AI performance avoid the initial safety decline entirely.
Psychological safety impact on performance is moderated by task type: safety levels impact performance strongly for ambiguous creative tasks (+1.8x performance at high safety) but barely impact performance for well-defined routine tasks (+0.2x at high safety). This suggests safety becomes more critical as AI augments cognitive work.
The relationship between safety and AI adoption is bidirectional: higher safety leads to faster adoption (r=0.62), but successful AI adoption also increases safety, creating a positive feedback loop. Teams breaking into this positive cycle achieved 3.4x better outcomes than teams stuck in the initial decline.
Impact & Application
Informs organizational change management for AI deployments. Adopted by 18 enterprises affecting 8K+ employees. Safety-focused change programs show 2x faster adoption and 1.8x higher sustained usage.
Contributors
Lead: Dr. Sofia Andersen (Human & Organizational Potential school). Collaborators from 12 enterprises across tech, finance, healthcare, and manufacturing. Supported by Gallup for validation of psychological safety measures.