Work Design and Multidimensional AI Threat as Predictors of Workplace AI Adoption and Depth of Use
Citation. Reich, A., Wolfe, D., Price, M., Choe, A., Kidd, F., & Wagner, H. (2026). Work Design and Multidimensional AI Threat as Predictors of Workplace AI Adoption and Depth of Use. arXiv:2602.23278.
Summary. This paper examines workplace AI adoption through the joint lens of work design and multidimensional AI threat. Using cross-sectional self-report survey data from 2,257 employees, it finds that structural job characteristics, especially skill variety and autonomy, are more consistent predictors of AI adoption than generalized AI threat, while different threat dimensions relate differently to depth of use. The practical claim is that AI enablement should not rely on training and messaging alone; it should attend to the opportunity structure of work, identity concerns, and possible workload expansion.
Key claims
- The study uses a cross-sectional organizational survey of employees (N = 2,257) and models AI adoption, frequency of use, and duration of use using four job-characteristics variables and four AI-threat dimensions.
- In logistic regression predicting AI adoption, skill variety was the strongest positive predictor (β = 0.31, p < .001, OR = 1.37, 95% CI [1.23, 1.52]), followed by autonomy (β = 0.20, p = .001, OR = 1.22, 95% CI [1.08, 1.36]).
- Loss of controllability had a small positive association with adoption (β = 0.11, p = .044, OR = 1.12), while task significance, feedback, work changes, skills/expertise threat, and status/position threat did not independently predict adoption once the full predictor set was included.
- For AI use frequency, skill variety was a positive predictor (β = 0.15, p < .001) and perceived work changes had a small positive association (β = 0.06, p = .049), while threat dimensions tied to controllability, status, and skills/expertise were not significant.
- For AI use duration, perceived work changes (β = 0.17, p < .001) and skill variety (β = 0.13, p = .001) predicted longer use, while task significance predicted shorter use (β = -0.10, p = .022) and status/position threat was directionally negative but marginal (β = -0.09, p = .062).
- The findings support treating AI threat as multidimensional rather than a single adoption barrier: work-change appraisals can covary positively with deeper use, while identity/status concerns may constrain sustained engagement.
Evidence & limitations
- Design: cross-sectional organizational survey with self-reported job characteristics, AI-threat appraisals, adoption, use frequency, and use duration.
- Evidence strength: useful associative evidence linking Job Characteristics Model constructs to workplace AI adoption and depth of use; the large sample and multivariable models make it stronger than anecdotal adoption claims.
- Limitations: not causal; adoption and use depth are self-reported; the paper does not test downstream outcomes such as performance, well-being, learning, skill development, or whether deeper AI use is adaptive versus maladaptive; the source is an arXiv preprint/whitepaper rather than peer-reviewed publication.