Trend Breakdown

The Evidence

What psychological shifts does managing alongside AI demand?

Managers supervising AI agents face documented psychological disruption: identity threat, fragile trust, and measurable harm to team psychological safety. The peer-reviewed literature confirms these costs. The prescribed remedies, broadly sold as AI leadership skills training, are largely untested. This breakdown scores what the evidence actually supports.

Published 4 Jun 2026 · 4 min read · 5 sources

Skip to the verdict ↓

Trend Science
Breakdown

Evidence-graded series

02What's being claimed

Managing AI agents demands a fundamentally different set of skills from managing human teams: as autonomous systems handle communications, delegate subtasks, and run multi-step workflows, managers must direct rather than supervise, calibrate trust in non-human collaborators, and govern decisions with no human counterpart. This is, the argument runs, a genuine leap in cognitive and relational demands.

The trend spread on three credibility rails: replicated organisational psychology findings gave it scientific grounding; consulting firms with large survey datasets created urgency; and rapid LLM-based agent deployment made the abstract immediately tangible for working managers. The underlying premise rests on genuine literature. Algorithm aversion, trust miscalibration, and identity disruption under AI automation are documented phenomena, not theorised ones ¹ ² ⁴.

As agentic systems take on decisions previously owned by humans, the manager's cognitive load, power relationships, and professional identity all shift simultaneously across individual, team, and organisational levels ³. Consulting research from McKinsey and BCG amplified this framing in 2025, positioning the transition as the decade's defining leadership challenge. The claim has genuine scientific grounding; the contention is whether prescribed solutions match the strength of evidence for that disruption.

Origin

AI agent theory

AI agent theory formalised autonomous goal-pursuit decades before agentic AI entered leadership vocabulary.

→

Vector

Andrew Ng / DeepLearning.AI

Ng's 2024 keynote translated autonomous LLM loops into practitioner language and seeded the leadership framing.

→

Spike

McKinsey & BCG 2025

Major consulting research positioned agentic AI as the decade's defining organisational challenge for leaders.

"Directing AI agents is a completely different skill set from managing people. As AI takes over more decisions, managers need to evolve fast, or they will find themselves obsolete in systems they nominally oversee."

- representative of the claim as it circulates online

03The evidence verdict

HiPerformance Culture The Evidence · Trend Breakdown

Verdict

Agentic AI creates real psychological disruption; prescribed leadership fixes lack experimental validation.

This trend lands here

Evidence for the claim: 52/100 (Moderate) 5 sources cited · 1 systematic multilevel review, 2 experimental studies, 1 time-lagged survey, 1 theoretical review · 2021-2025 · how we grade ↓

Every line below is graded by the strength of the evidence behind it.

Gold: strongest available evidence
Silver: solid, not decisive
Bronze: preliminary or indirect

how we grade ↓

✓

What holds up

✓

Managers experience measurable identity disruption and competence threat when AI systems automate decisions they previously owned.¹

GoldSupporting study: Leyer 2021

✓

Cognitive trust in AI teammates forms faster than in human teammates but is significantly more fragile; AI errors damage team trust disproportionately compared to equivalent human errors.²

GoldSupporting study: Georganta 2024

✓

AI integration reshapes managerial role identity, cognitive load, and power structures simultaneously across individual, team, and organisational levels.³

SilverSupporting study: Bankins 2023

What doesn't

Algorithm aversion is paradoxically strongest in high-stakes decisions; managers resist AI assistance most precisely where superior AI performance would deliver the greatest benefit.⁴

GoldSupporting study: Filiz 2023

AI adoption significantly reduces employee psychological safety and increases depression risk; 'empowerment through AI' narratives systematically understate these documented mental health costs.⁵

Safety-criticalGoldSupporting study: Kim 2025

04So what do you actually do

Where the evidence holds, it supports a specific, targeted approach.

Broad AI leadership upskilling programmes may help, but these items rest on the strongest available evidence.

01Keep human decision authority in high-stakes choices rather than fully automating them.

02Calibrate AI trust explicitly: neither dismiss AI recommendations in complex decisions nor defer to them uncritically.

03Treat psychological safety as a first-order concern when deploying AI alongside existing teams.

04Build explicit team-level protocols for when and how AI agent errors are handled; affective trust fractures disproportionately after AI mistakes.

05The studies

Tier inputs Design quality Measurement precision Causal clarity Replication value

GoldThe strongest evidence available for the specific line it supports: typically systematic reviews, meta-analyses, or randomised trials, or definitive mechanistic work where the line is a mechanistic claim.

SilverSolid but not decisive support: controlled or observational studies, or direct evidence that is small or not yet replicated.

BronzePreliminary or indirect signals; suggestive, never sufficient on their own.

A per-source tier is displayed only when all four axis scores and the weights and thresholds that compute the tier are recorded. Sources without those inputs remain unscored; no tier is inferred from a design label, sample-size phrase, or existing editorial chip. Every source still discloses its design and a resolvable DOI.

Conceptual review with managerial decision vignettes

Leyer & Schneider Business Horizons · 2021

Managers face an identity threat when AI automates decisions they previously owned. The paper frames neither over-reliance nor under-reliance on AI recommendations as neutral: both create distinct organisational risks. Metacognitive skills to supervise AI are proposed as the critical buffer, yet no RCT validates this prescription.

doi:10.1016/j.bushor.2021.02.026 Verify ↗

N=828 participants across two controlled team experiments

Two experiments · N=828

Georganta & Ulfert Journal of Occupational and Organizational Psychology · 2024

Cognitive trust in AI teammates emerged faster than in human teammates, but affective trust was systematically lower and more volatile. Conventional trust-building frameworks designed for human reciprocity do not transfer cleanly to AI agents. Teams require explicit new trust management protocols when AI agents are introduced.

doi:10.1111/joop.12504 Verify ↗

Systematic multilevel review

Bankins et al. Journal of Organizational Behavior · 2023

A multilevel review identifies AI literacy as a psychological buffer for managers and calls for new theory on how managers adapt when direct reports include AI agents. The review covers individual, team, and organisational levels, documenting consistent evidence of identity disruption, cognitive load changes, and power-structure shifts.

doi:10.1002/job.2735 Verify ↗

50.7% chose AI for serious decisions despite superior AI accuracy

Economic experiment · n=143

Filiz et al. PLOS ONE · 2023

Algorithm aversion was most pronounced in high-stakes decisions: only 50.7% of participants chose the algorithm for serious choices despite it demonstrably outperforming human experts (70% vs 60% accuracy). For trivial decisions, 70.83% chose the algorithm. Managers resist AI assistance most precisely where its superior performance would matter most.

doi:10.1371/journal.pone.0278751 Verify ↗

3-wave longitudinal design establishing causal order · n=381

3-wave time-lagged survey · n=381

Kim et al. Humanities and Social Sciences Communications · 2025

AI adoption significantly reduced employee psychological safety, which in turn increased depression in a 3-wave longitudinal design establishing causal order. Ethical leadership moderated the harm, suggesting the manager's relational role becomes more critical when AI enters the team. Organisations deploying AI without attending to psychological safety face measurable mental health risks.

doi:10.1057/s41599-025-05040-2 Verify ↗

06The verdict triad

Claim

A Qualitatively Different Challenge

Directing AI agents creates distinct psychological demands on trust, identity, and decision authority that differ qualitatively from supervising human teams. The evidence maps three simultaneous disruptions: managers lose ownership of decisions they previously held, must calibrate trust in non-human collaborators, and navigate shifted power structures. These are not adaptations of existing skills; they are categorically new ones.

Consequence

Uncalibrated Adoption Produces Documented Harms

Without calibration, two failure modes compound. Managers resist AI recommendations most in precisely the high-stakes decisions where AI demonstrably outperforms them. Simultaneously, teams report reduced psychological safety and increased depression risk. AI trust, which forms quickly, fractures disproportionately at the first error. Neither problem resolves passively over time.

Lever

Targeted Calibration Over Blanket Upskilling

Three adjustments have the strongest evidential footing. Preserve human decision authority in high-stakes choices, where algorithm aversion is paradoxically strongest. Build explicit protocols for how AI errors are handled in team settings before affective trust breaks. Treat the manager's relational, supportive role as a primary buffer against AI-driven psychological safety erosion.

07The protocol

08What to do next

What to do next

Is your leadership approach ready for agentic AI?

The HPC Leadership Assessment maps your trust calibration, decision authority frameworks, and psychological safety practices against the evidence. Take it before rolling out AI agents in your team.

Take the assessment

09Share & references

Update log

4 Jun 2026Initial publication. 5 sources · 2021-2025.

20 Aug 2026Machine layer added: structured claim data with an authored veracity verdict, accessible meter with stated evidence position, grading disclosure, and related links and calls to action wired to live destinations.

Next reviewDue by February 2027, or earlier if a new controlled trial or systematic review lands.

Bibliography · every source, resolvable

01Leyer, M. & Schneider, S. (2021). Decision augmentation and automation with artificial intelligence: Threat or opportunity for managers?. Business Horizons, 64(5), 711-724. doi:10.1016/j.bushor.2021.02.026 Verify ↗

02Georganta, E. & Ulfert, A. (2024). Would you trust an AI team member? Team trust in human– AI teams. Journal of Occupational and Organizational Psychology, 97(3), 1212-1241. doi:10.1111/joop.12504 Verify ↗

03Bankins, S., Ocampo, A.C., Marrone, M., Restubog, S.L.D. & Woo, S.E. (2023). A multilevel review of artificial intelligence in organizations: Implications for organizational behavior research and practice. Journal of Organizational Behavior, 45(2), 159-182. doi:10.1002/job.2735 Verify ↗

04Filiz, I., Judek, J.R., Lorenz, M. & Spiwoks, M. (2023). The extent of algorithm aversion in decision-making situations with varying gravity. PLOS ONE, 18(2), e0278751. doi:10.1371/journal.pone.0278751 Verify ↗

05Kim, B., Kim, M. & Lee, J. (2025). The dark side of artificial intelligence adoption: linking artificial intelligence adoption to employee depression via psychological safety and ethical leadership. Humanities and Social Sciences Communications, 12(1). doi:10.1057/s41599-025-05040-2 Verify ↗