AI improves workplace efficiency but fuels employee anxiety; frontline staff face the greatest displacement risk, while firm-level training and transparent algorithm governance substantially reduce job‑insecurity effects.

Main Finding

AI adoption in firms produces dual effects: it raises employee and organizational productivity by automating routine tasks, while simultaneously increasing employees’ perceptions of job insecurity and related negative psychological outcomes. These effects are heterogeneous across organizational levels (frontline workers most negatively affected), and organizational policies—particularly AI training and transparent algorithm governance—meaningfully mitigate the negative impacts.

Key Points

• Evidence base: systematic review of 186 peer‑reviewed articles published Jan 2024–May 2026 (Chinese and international journals).
• Broad consensus:
  • Positive: AI (both analytical AI and generative AI/LLMs) automates repetitive tasks, freeing time for higher‑value work and improving productivity.
  • Negative: AI increases job insecurity, career anxiety, and can erode some independent problem‑solving skills if over‑relied upon.
• Heterogeneity by role:
  • Frontline employees: highest risk of displacement and strongest negative psychological effects.
  • Middle managers: mixed effects—support for decision‑making but some automation of managerial tasks.
  • Senior managers: largely benefit from decision support; little increase in job insecurity.
• Mechanisms:
  • Mediation: job insecurity is a key mediator linking AI adoption to reduced job satisfaction and poorer psychological well‑being (supported by ~69 studies with mediation models).
  • Moderation: organizational support (AI training, transparent algorithm governance) weakens the link between AI adoption and job insecurity (~56 studies consider this).
• Descriptive stats from the review:
  • Papers by year: 62 (2024), 79 (2025), 45 (Jan–May 2026).
  • Analytical level: 41 macro, 72 organizational behavior, 73 individual psychology.
  • Industry coverage: manufacturing (61), IT (58), finance (54), cross‑industry (13).
  • Topics: 127 on efficiency gains, 142 on job insecurity, 56 on organizational support.
• Divergent/limited findings: some (few) studies argue negative effects decline as employees adapt to human–AI collaboration, but robust longitudinal evidence is limited.

Data & Methods

• Method: systematic literature review (retrieval → screening → coding → thematic synthesis).
• Databases searched: CNKI (CSSCI, PKU Core), Web of Science Core Collection, Frontiers journals, and leading HR / labor economics journals.
• Period: Jan 2024 – May 2026.
• Inclusion: peer‑reviewed journal articles focusing on enterprise AI and employee outcomes (work performance or psychological well‑being), organizational or individual level.
• Exclusion: pre‑2024 work, macro‑only labor market studies without organizational focus, purely technical/algorithmic papers, non‑peer‑reviewed sources.
• Theoretical framing: integrates Task Automation Theory, low‑cost prediction view of AI, and the General‑Purpose Technology (GPT) perspective. Three formal propositions:
• AI raises efficiency while increasing perceived job insecurity.
• Job insecurity mediates AI’s effect on work outcomes and well‑being.
• Organizational AI training and algorithmic transparency moderate (reduce) AI→job insecurity.

Implications for AI Economics

• Labor demand composition: empirical literature reinforces that AI substitutes routine tasks and complements non‑routine cognitive work—models of labor demand should explicitly incorporate task‑level substitution/complementarity and heterogeneous worker impacts.
• Productivity vs displacement trade‑off: productivity gains coexist with heightened job insecurity; welfare and policy analyses must weigh aggregate gains against distributional and psychological costs to workers, especially lower‑status workers.
• Role of firm‑level governance: firm investments in worker training and transparent algorithmic governance are empirically important moderators. Economic models and empirical work should include firm‑level governance and complementary capital as mediating factors when estimating AI’s net labor market effects.
• Measurement and identification:
  • Researchers should prioritize longitudinal and causal designs (to distinguish short‑term anxiety from persistent displacement).
  • Incorporate direct measures of algorithmic transparency, training intensity, and task reallocation to improve identification of mechanisms.
• Policy relevance:
  • Targeted training and re‑skilling programs at the firm or policy level can reduce adverse worker outcomes and ease transitions.
  • Labor market policies (retraining subsidies, portable benefits, transition assistance) should be informed by heterogeneous exposure—frontline workers are highest priority.
• Future research priorities for AI economics:
  • Longitudinal studies on adaptation and dynamics of job insecurity.
  • Causal estimates of how specific organizational practices (e.g., transparency rules, training curricula) change labor outcomes.
  • Distributional impacts on wages, hours, and career progression across firm sizes, sectors, and countries.
  • Modeling AI as a GPT that requires complementary investments and institutional adjustments to capture firm‑level heterogeneity in outcomes.

Limitations noted by the review: reliance on secondary studies (no primary new empirical data), focus on 2024–2026 peer‑reviewed literature (possible publication lag and selection), and limited longitudinal causal evidence.