Return-to-office edicts, rushed AI pilots, and rolling reorganizations have left managers holding the bag: new mandates from above, anxious teams below, and little cover if experiments fail. The numbers show the strain. Global employee engagement fell to 20 percent in 2025, and manager engagement slid from 27 to 22 percent in a single year, part of a nine-point drop since 2022, with Gallup estimating the global productivity cost at roughly 10 trillion dollars, about 9 percent of world GDP (Gallup, 2025). When managers are worried about being wrong, they do the rational thing in a punitive system. They narrow discretion, stop asking hard questions, and pass risk down. That is how learning dies.
What psychological safety changes in the room
Psychological safety is not a slogan. It is a team-level belief that it is safe to take interpersonal risks, such as admitting a mistake or challenging a plan (Edmondson, 1999). In a field study of 51 work teams in a manufacturing company, teams with higher psychological safety showed more learning behavior, including asking for help and discussing errors. Team efficacy, or confidence in the team’s abilities, did not predict those behaviors once psychological safety was accounted for. The path to performance ran through learning: learning behavior mediated the relationship between psychological safety and team outcomes (Edmondson, 1999). This is not a lab result and it is not a universal law. The evidence is from one firm and is cross-sectional. The practical lesson is still sharp. If people expect interpersonal risk to be punished, they will avoid the very behaviors that surface problems early.
That mechanism matters in 2025 because so much work is hypothesis-driven. AI use cases are unproven, customer needs shift quickly, and hybrid coordination is fragile. In such settings, performance depends on noticing anomalies and sharing partial information fast. That is precisely the pattern psychological safety enables.
Managers control the levers, not the weather
The broader evidence base shows that psychological safety travels with the behaviors firms claim to want. A meta-analysis aggregating 136 independent samples found strong positive corrected correlations with learning behavior (0.62), information sharing (0.52), work engagement (0.45), and task performance (0.43). The link with creativity was much smaller, at 0.13 (Frazier et al., 2017). The same review points to design levers that predict psychological safety: autonomy (0.47), role clarity (0.63), job interdependence, and a supportive context (0.49). One estimate for interdependence at the individual level is very large, 0.86, but it rests on only five samples and should be treated with caution. The authors report smaller effects at the group level. The managerial reading is straightforward. You do not create safety with pep talks. You create it by clarifying who decides what, by granting latitude where it helps, by making interdependence explicit, and by surrounding teams with visible support.
This also reframes the engagement crash. Gallup reports that in best-practice organizations, 79 percent of managers are engaged, nearly four times the global average of 22 percent (Gallup, 2025). Surveys cannot prove causation, but the gap is a design signal, not a mood swing. Where the job is set up to enable voice and learning, managers stay engaged. Where the job is defined as policing attendance and absorbing blame, they disengage.
Flatten, but do not abdicate
Some leaders respond to uncertainty by stripping layers and asking teams to self-organize. That can help, but only if authority is redistributed by design, not removed by hope. Self-managing organizations decentralize authority radically and eliminate the manager-subordinate relationship as a formal feature. They can be adept at solving cross-functional, nonroutine problems that stump traditional hierarchies, which are optimized for reliable execution in stable contexts and tend to reinforce status differences that suppress development (Lee and Edmondson, 2017). The empirical base on outcomes is still limited, and this is a theoretical synthesis, not a randomized trial, so do not overread it. The operational point travels well across structures. Whether you keep hierarchy or flatten, complex work crosses boundaries, exposes fallibility, and threatens status. Psychological safety is the lubricant that lets teams navigate those frictions without shutting down.
In a self-managing setting, people must raise interdependencies and constraints early because no single boss coordinates by fiat. In a conventional setting facing novel work, the same is true because no one can know enough in advance. The practices look similar: frame tasks as learning problems, normalize help-seeking, and make it routine to test and revise. Those are managerial choices, even when you keep managers.
A practical upstream agenda
If your managers are the bottleneck, fix the job they do.
-
Specify where learning beats certainty. For initiatives with real unknowns, state hypotheses and expected failure modes at the start. Invite disconfirming views in the first meeting. This turns critique from personal risk into part of the job, aligning with how psychological safety fuels learning behavior on the way to better outcomes (Edmondson, 1999).
-
Redesign for clarity and latitude. Publish decision rights and interfaces so people know when to escalate and when to act. Grant autonomy where speed matters, paired with tight feedback loops. These are antecedents associated with higher psychological safety across studies, with meaningful correlations for autonomy and role clarity, and with supportive context effects that managers can create day to day (Frazier et al., 2017).
-
Make interdependence visible and bounded. Map handoffs and shared risks so that asking for help is expected. Treat the size of the interdependence effect cautiously, since some estimates rest on few samples, and aim for clear, tested routines rather than vague calls for collaboration (Frazier et al., 2017).
-
Shield and amplify. When a team surfaces an error early, respond by fixing the process first, not the person. Escalate their learning to peers so that voice has visible payoff. This is how a supportive context becomes real, not rhetorical.
None of this requires a new slogan. It does require aligning incentives and rituals with the evidence that safety drives learning behavior, and that learning behavior is the path to performance when the work is not fully knowable in advance (Edmondson, 1999). It also requires taking manager engagement as a design variable, not a character flaw. If those who rate performance and set norms feel safe to say “I do not know yet,” their teams will say it sooner too. That is how you turn a frightening year into a teachable one, and how you move engagement from a survey result to an operating choice (Gallup, 2025; Frazier et al., 2017; Lee and Edmondson, 2017).
Sources
- Edmondson (1999). Psychological Safety and Learning Behavior in Work Teams, Administrative Science Quarterly. Defines team psychological safety and finds it relates to learning behavior, which mediates its link to performance in a field study of 51 teams. https://journals.sagepub.com/doi/10.2307/2666999
- Frazier, Fainshmidt, Klinger, Pezeshkan and Vracheva (2017). Psychological Safety: A Meta-Analytic Review and Extension, Personnel Psychology. Aggregates 136 samples showing strong correlations between psychological safety and learning, information sharing, engagement, and task performance, and identifies job-design antecedents with caveats on interdependence estimates. https://doi.org/10.1111/peps.12183
- Lee and Edmondson (2017). Self-managing organizations: Exploring the limits of less-hierarchical organizing, Research in Organizational Behavior. Theoretical synthesis defining SMOs and arguing where hierarchy helps and where it hinders complex problem solving and development. https://www.sciencedirect.com/science/article/abs/pii/S0191308517300059
- Gallup (2025). State of the Global Workplace 2025. Reports 20 percent global engagement, a sharp drop in manager engagement to 22 percent, and a productivity cost estimate of about 10 trillion dollars, highlighting the design gap with best-practice organizations. https://www.gallup.com/workplace/659610/state-global-workplace.aspx