Artificial Intelligence and Job Design: Insights from Job Demands–Resources Theory
Citation. Bakker, A. B., Junker, T. L., Scharp, Y. S., & Cerne, M. (2026). Artificial Intelligence and Job Design: Insights from Job Demands–Resources Theory. Technology in Society, 103533. https://doi.org/10.1016/j.techsoc.2026.103533
Summary. This integrative review applies Job Demands–Resources theory to AI-enabled job design. It argues that AI can increase or decrease demands and resources, depending on whether adoption is top-down or bottom-up and whether use is automation- or augmentation-oriented. The authors distinguish AI-readiness job resources from personal resources and propose reciprocal gain and loss spirals through which job design, well-being, AI readiness, and future AI use may reinforce one another.
Key claims
- AI adoption should not be treated as a uniform intervention: the authors’ 2 × 2 frame separates adoption origin (top-down versus bottom-up) from use type (automation versus augmentation), and argues that these configurations have different implications for demands, autonomy, resources, and well-being.
- AI can reduce first-order routine or cognitive demands while substituting them with more complex escalation, review, accountability, emotional, or creative work; apparent workload relief is therefore a reconfiguration claim, not evidence of net job-quality improvement.
- AI can function as a cognitive resource and feedback source, but top-down algorithmic management can erode autonomy, task completeness, skill-development opportunities, and social resources; the review treats the direction of effect as contingent on adoption configuration and workplace conditions.
- AI-readiness job resources (for example, training, time to experiment, supportive leadership, psychological safety, clear responsibility allocation, and organizational support) and AI-readiness personal resources (for example, AI self-efficacy, digital literacy, metacognitive skill, domain expertise, and learning orientation) are distinct intervention targets.
- The proposed dynamic JD–R model predicts recursive gain and loss spirals: resource-rich work can support AI learning, engagement, and further resource accumulation, whereas constrained work can make AI a demand that weakens readiness and widens job-design inequality over time.
- The authors call for validated measures of adoption configuration, episode-level and longitudinal designs, and tests of reciprocal effects among AI readiness, job resources, demands, engagement, strain, and job crafting.
Evidence & limitations
- This is an integrative theory review and research agenda, not a systematic review, meta-analysis, or new empirical study; its central 2 × 2 and dynamic-model claims are organizing propositions rather than estimates of causal effects.
- The review draws on heterogeneous studies across occupations, countries, technologies, and research designs. Individual results it cites should be read in their original studies rather than aggregated as a common effect size.
- The ingested file is Elsevier’s 27 August 2026 journal pre-proof, explicitly described in the PDF as not yet the definitive version of record; Crossref supplies the public journal-article DOI, August 2026 issue date, journal, article number, and publisher provenance. The Drive binary SHA-256 is recorded above.
- The article declares ChatGPT-5 was used to improve manuscript readability and language; this is an authorship disclosure, not model-directed source text. Injection scan found no instructions aimed at the ingesting agent.