Return-to-office edicts, AI anxiety, and talk of the great flattening have managers second-guessing their own roles. The numbers are not reassuring. Global employee engagement fell to 20 percent in 2025, and the share of engaged managers dropped to 22 percent from 27 percent in a single year, part of a trend that Gallup estimates costs about 10 trillion dollars in lost productivity, or 9 percent of world GDP (Gallup, 2025). If fewer managers are expected to coordinate more complex, hybrid work, voice and initiative must travel farther with less supervision. That only happens if people feel safe to surface problems before they metastasize.
What psychological safety actually predicts
Psychological safety is not a pep talk. It is a team-level belief that the group is safe for interpersonal risk taking, such as admitting an error or asking for help. The construct and its consequences were established in a field study of 51 teams inside a single manufacturing firm. In that study, psychological safety was associated with learning behaviors such as seeking feedback, experimenting, and speaking up. Team efficacy was not, once psychological safety was taken into account. Learning behavior mediated the link between safety and performance, which implies that safety matters because it gets people to do the unglamorous work of improvement in the open. The study is cross-sectional and limited to one company, so the mechanism should be read as pattern, not proof of causation, but it remains the seminal anchor of the field (Edmondson, 1999).
A decade and a half of subsequent research largely confirms the practical relationship. A comprehensive meta-analysis that pooled 136 independent samples covering more than 22,000 people and almost 5,000 groups found strong corrected correlations between psychological safety and learning behavior (ρ̂ = .62), information sharing (ρ̂ = .52), engagement (ρ̂ = .45), and task performance (ρ̂ = .43). The link to creativity was smaller (ρ̂ = .13). On the antecedent side, job characteristics correlated with safety, including autonomy (ρ̂ = .47), interdependence (ρ̂ = .86 with an important caveat that this large estimate at the individual level came from only five samples, while the group-level estimate was .40), role clarity (ρ̂ = .63), and a supportive context (ρ̂ = .49). These are correlations, not causal effects, but across diverse settings the pattern is consistent: design the work so people must coordinate, understand their role, and have room to act, and psychological safety is more likely to emerge alongside the learning behaviors leaders say they want (Frazier et al., 2017).
Why flattening raises the stakes
The structure of authority shapes whether people speak up. A synthesis of research on self-managing organizations defines them as systems that decentralize authority in a radical, formal, systematic way and remove the conventional boss–subordinate tie. The same review argues that managerial hierarchy is effective for reliable execution in stable environments, but tends to inhibit complex, cross-boundary problem solving and reinforce status differences that stunt development. In other words, when work becomes less routine and coordination demands rise, the liabilities of hierarchy grow more visible. None of this is a plea for anarchy. It is a reminder that as formal control recedes, the operating system must replace it with conditions that make initiative safe and error-reporting routine (Lee and Edmondson, 2017).
Hybrid work and AI adoption bring the same design problem to more conventional firms. Decisions span time zones and tools, status markers blur on a video grid, and people hesitate to reveal uncertainty about new technologies. Meanwhile, the engagement drop among managers should be read as a system warning light. In best-practice organizations, 79 percent of managers are engaged, almost four times the global average. That gap suggests the difference is organizational design, not the business cycle (Gallup, 2025). If managers are less present or spread thinner, the work must carry more of the managerial load. Psychological safety is what allows the work to carry it without silence and rework.
Design it into the work, not the wall art
The research points to concrete levers. First, structure interdependence and role clarity. Many teams today are interdependent in name only. Explicitly map who needs what from whom, by when, and in what format. The meta-analysis links both interdependence and role clarity to psychological safety, with the important caveat that the very large interdependence estimate at the individual level rests on few samples, while the group-level relationship is more modest. Either way, making dependencies visible raises the odds that people ask for help before deadlines make it expensive (Frazier et al., 2017).
Second, treat learning behavior as the output to manage. In Edmondson’s field study, learning behavior sat between safety and performance. If you do not see pilots, debriefs, or fast feedback loops, you should not expect gains in quality or speed. Put time on the calendar for problem-solving and post-mortems, and make near-misses discussable without blame. The original evidence comes from one company with a cross-sectional design, so avoid magical thinking. The practical read is that behaviors are observable and coachable in any context (Edmondson, 1999).
Third, shift managers from judges to context designers. Where the organization is flattening, managers’ scarcest resource is attention. Use it to build the conditions that the evidence associates with safety: autonomy where possible, clarity where essential, and a supportive context that rewards speaking up. The synthesis on self-managing organizations is clear that hierarchy excels at repeatable execution but struggles with nonroutine, cross-functional challenges. If that is your work, overinvest in routines that invite dissent and surface interdependencies early (Lee and Edmondson, 2017).
Finally, measure what is moving. Engagement is an outcome that boards track, and the global fall suggests why psychological safety has become a CEO topic. But do not confuse a climate score with a capability. Track leading indicators such as the volume of documented experiments, the rate of issue-raising before launch gates, or the number of process changes triggered by frontline suggestions. If those are flat while engagement surveys improve, you are treating safety as sentiment, not system. The firms that show much higher manager engagement levels than the global average demonstrate that design choices can move the dial at scale, even if surveys cannot tell you which lever did the work (Gallup, 2025).
The through-line is simple. Psychological safety is not about being nice. It is about removing the personal threat in the very behaviors that make organizations smarter: asking, testing, correcting. The classic evidence ties safety to learning behavior and performance at the team level. The broader synthesis shows consistent links across industries and job designs. And the structural argument warns that as hierarchy recedes or work becomes more complex, the need for safety grows. Treat it as a system property you can design, then hold yourself to evidence that the system is doing what you built it to do.
Sources
- Edmondson (1999). Psychological Safety and Learning Behavior in Work Teams, Administrative Science Quarterly. Defines team psychological safety and shows in a field study of 51 teams that it relates to learning behaviors which mediate its link to performance, with team efficacy falling away when safety is controlled. 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 to show strong correlations between psychological safety and learning behavior, information sharing, engagement, and task performance, and identifies job design antecedents with noted 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. Defines self-managing organizations and synthesizes evidence that hierarchy suits reliable execution but inhibits complex cross-boundary problem solving, underscoring the need for conditions that support voice. https://www.sciencedirect.com/science/article/abs/pii/S0191308517300059
- Gallup (2025). State of the Global Workplace 2025. Reports global engagement at 20 percent, a drop in manager engagement from 27 to 22 percent, and an estimated 10 trillion dollar productivity cost, illustrating the scale and design stakes for psychological safety. https://www.gallup.com/workplace/659610/state-global-workplace.aspx