The research, led by Azatullah Zaheer of Salam University in Kabul together with colleagues from Malaysia, Bangladesh, and the United Arab Emirates, surveyed 404 employees working in organizations in Kabul, Afghanistan, that have adopted generative AI tools. The choice of setting is significant. Most studies of workplace AI adoption have been conducted in wealthy economies with mature digital infrastructures, leaving a substantial gap in our understanding of how these technologies perform in developing contexts, where connectivity may be patchy, training resources scarce, and organizational support uneven. By focusing on Kabul, the team offers one of the first empirical windows into how generative AI reshapes productivity where digital transformation is still very much a work in progress.

The study rests on two well-established theoretical pillars. The first is Task–Technology Fit Theory, which holds that technology only improves performance when its capabilities match the demands of the tasks at hand. The second is Social Cognitive Theory, which emphasizes that human behavior is shaped by an interplay of personal capabilities, beliefs, and environmental factors. Combining these frameworks, the researchers conceptualized generative AI adoption not as a single yes-or-no variable but as a multidimensional construct with three distinct dimensions: frequency, meaning how often employees use AI tools; depth, meaning how intensively and sophisticatedly they integrate AI into their workflows; and purpose, meaning the range of tasks for which they deploy it.

To test the relationships among these dimensions, digital skills, AI trust, and productivity, the team employed structural equation modeling, a statistical technique that allows researchers to estimate networks of direct and indirect relationships among observed and latent variables simultaneously. This approach is particularly well suited to questions like the present one, where the pathway from technology adoption to productivity is unlikely to be a simple straight line. The researchers hypothesized that digital skills — the ability to operate, adapt, and troubleshoot digital tools — would act as a mediator, carrying the effect of AI adoption through to measurable productivity gains. They also tested whether trust in AI would moderate, or amplify, the entire chain of relationships.

The results delivered a clear verdict on the first two dimensions. Both the frequency and the depth of generative AI adoption were positively and significantly associated with employee productivity. In other words, workers who used AI tools more often, and who wove them more thoroughly into the fabric of their daily tasks, tended to report higher productivity. This aligns neatly with Task–Technology Fit Theory: the more a tool is used, and the more deeply it is matched to actual task demands, the more performance benefits it delivers. Interestingly, the third dimension — the purpose of AI adoption — showed no significant direct association with productivity. Simply using AI for a wider variety of tasks did not, by itself, translate into better outcomes, suggesting that breadth of use without frequency or depth may spread attention thin rather than compound gains.

The mediation findings may be the most consequential part of the study. Digital skills demonstrated significant indirect associations between generative AI adoption and employee productivity. In plain terms, AI adoption did not magically make workers more productive on its own; rather, its benefits flowed through the workers’ own capabilities. Employees with stronger digital skills were better positioned to convert their AI usage into tangible output, whether that meant crafting more effective prompts, verifying and refining machine-generated content, or integrating AI outputs into larger workflows. This finding reframes the popular narrative around AI and productivity: the technology is less a replacement for human skill than an amplifier of it, and the amplifier only works when there is a signal worth boosting.

Perhaps the most surprising result was the one that failed to materialize. The researchers hypothesized that trust in AI would moderate the relationship between adoption and productivity — that employees who trusted the technology would extract more value from it. The data did not support this. AI trust did not significantly moderate these relationships, a finding that cuts against a substantial body of adoption literature in which trust is treated as a gating factor for technology acceptance. The authors’ result hints that in the workplace context, what matters may not be whether employees emotionally trust AI but whether they possess the competence to use it well. Trust, in this framing, may be a prerequisite for adoption rather than a multiplier of its benefits — or it may simply matter less than skills once tools are already in daily use.

The practical implications reach well beyond Kabul. For managers contemplating generative AI rollouts, the study suggests that procurement alone is a poor strategy. Buying licenses and granting access addresses only the frequency dimension; without investment in training that builds depth of use and strengthens digital skills, organizations may see adoption statistics rise while productivity remains flat. The finding that purpose of use showed no direct effect adds a caution against broad, unfocused AI mandates. Instead, the evidence points toward targeted skill development: teaching employees not just how to operate AI tools but how to embed them deeply and appropriately into specific, recurring tasks where the technology’s strengths match the work’s demands.

The study also carries a developmental dimension. The authors explicitly connect their findings to the United Nations Sustainable Development Goals, particularly SDG 8 on decent work and economic growth and SDG 9 on industry, innovation, and infrastructure. In economies like Afghanistan’s, where formal employment is fragile and digital infrastructure is still maturing, the difference between AI as a productivity engine and AI as an expensive distraction could have outsized consequences. The research suggests that digital skills training may be among the highest-leverage interventions available, allowing workers in developing contexts to leapfrog into AI-augmented productivity rather than being left behind by a technological wave they cannot ride.

As with any cross-sectional survey, the study has limits that invite further work. Measuring adoption and productivity at a single point in time cannot establish causality with certainty, and self-reported productivity may diverge from objective output measures. The single-city sample, while valuable for illuminating an understudied context, raises questions about generalizability to other developing and developed settings alike. Still, the core message is likely to travel. Generative AI is not a productivity switch that organizations can simply flip. Its benefits are conditional — conditional on frequent use, conditional on deep integration, and above all conditional on the human skills that turn machine output into meaningful work. In the emerging economy of AI-augmented labor, it appears, the decisive investment is not in the machines but in the people who use them.

Subject of Research: The relationship between generative AI adoption dimensions, digital skills, AI trust, and employee productivity in developing-country workplaces

Article Title: Generative AI utilization and employee productivity through digital skills with the moderating role of AI trust

Article References: Generative AI utilization and employee productivity through digital skills with the moderating role of AI trust. (n.d.). https://doi.org/10.1007/s43621-026-04863-6

Image Credits: AI Generated

DOI: 10.1007/s43621-026-04863-6

Keywords: generative AI, employee productivity, digital skills, AI trust, workplace innovation, Task-Technology Fit Theory, Social Cognitive Theory, structural equation modeling, Afghanistan, developing economies, sustainable development, digital transformation