As AI and automation technology rapidly redefine job tasks and workplace structures, my research investigates their effects on students, workers, and firms. Using innovative experiments and large-scale data, I trace behavioral and strategic adjustments through which technological change reshapes educational and occupational pathways. My findings show that digital advances are influencing the labor-market behavior of today’s workers, the career choices of tomorrow’s workforce, as well as firms’ future training strategies.

Individual and Firm Economics of Automation and AI

The emergence of generative AI tools like ChatGPT has fundamentally altered the automation landscape, now threatening occupations previously considered immune to technological substitution. Understanding how they already shape choices on both sides of the market is essential for averting skill shortages and ensuring equitable access to future jobs.

Exploiting ChatGPT’s surprise launch in November 2022, in a first project we deliver the first causal evidence that generative AI (GenAI) redirects occupational choices. A difference-in-discontinuity design applied to 45 million Swiss apprenticeship-search queries uncovers an 8 % and persistent drop in the supply of apprentices, strongest in occupations with a large share of cognitive tasks, high demands on language skills, and those previously considered “safe” from automation. Alongside evidence of declining applicant quality in Switzerland’s most popular—and highly AI-exposed—commercial apprenticeship, the project suggests that teenagers' perceptions of generative-AI risks, mirrored in their occupational choices, closely align with those of labor-market experts.

In another project, we demonstrate that individuals place substantial monetary value on reducing their risk of automation. In a discrete-choice experiment with nearly 6,000 participants, we found that respondents would accept salary reductions of almost 20% of the median annual wage to reduce their automation risk by 10 percentage points. This finding reveals the profound anxiety surrounding AI-driven job displacement and suggests that automation concerns significantly influence career decisions. Notably, willingness to pay for risk reduction increases with higher baseline automation risk levels, contrary to patterns observed in other risk contexts. Additionally, men, risk-tolerant, younger, and tertiary-educated individuals put relatively less value on reducing the risk of automation.

My examination of routine‐job trends in Switzerland from 1992 to 2018 offers essential context for today’s AI‐driven disruptions. Analyzing large-scale panel data, I found that the routine employment declined gradually because fewer workers entered such jobs, rather than because exits (e.g. layoffs) rose, and resulted in a general upgrading of the labor force instead of job-polarization. Switzerland’s pre-GenAI digitalization, therefore, affected workers less disruptively than the pattern observed in economies like the United States.

Building on these insights, my current work with survey data investigates labor demand, occupational preferences, and skill formation in the context of AI: how expected automation changes firms’ demand for apprentices; how teenagers trade off wage, GenAI use, automation risk and lifelong learning in their future occupations; and which factors affect young students’ AI engagement patterns in school and home contexts.

By conducting a vignette experiment with 2,890 firms, we find that a 10-percentage-point rise in the share of a trained worker’s tasks projected to be automated lowers intended apprentice positions by 1.51 points, an effect that weakens when the adoption horizon is delayed. The pull-back concentrates on routine-intensive and AI-exposed occupations and among larger firms.

In a discrete-choice experiment with over 7,000 adolescents, initial findings suggest that prospective workers assess the combination of workplace generative AI usage and automation risk as particularly dispreferable, or “scary,” and that they are willing to accept lower wages for occupations that require continuous education to work well and keep up with technology. Teenagers assess GenAI usage differently based on their gender and track choice, while general education students put less value on continuous education and lower automation risk.

Through survey data from 5,000 Swiss students aged 8-18, our work reveals significant contextual differences in AI engagement patterns, from passive homework assistance to interactive collaborative problem-solving. The study aims to identify distinct AI-user typologies related to digital infrastructure, sociodemographic factors, and teacher practices, to better understand how individual resources and institutional settings shape AI adoption in an increasingly automated economy.

Education Credentials in the Labor Market

My second research focus examines how individuals value educational credentials in an era of rapid technological change. This work challenges conventional assumptions about education as an intrinsic good, instead revealing that individuals approach educational decisions through conscious or unconscious cost-benefit calculations.

Through discrete-choice experiments, we investigate preferences for different educational pathways when controlling for wages, hierarchical status, and automation risk. Initial findings suggest that respondents treat higher education primarily as an economic investment rather than as a pursuit for intrinsic fulfillment. This economic rationality extends to incomplete degrees, where my study with HR recruiters reveals that university dropouts from majors unrelated to the job to be filled face substantially lower labor market outcomes than those who never enrolled, while alternative forms of human capital accumulation, like internships, are significantly preferred.

In a forthcoming study, I link Swiss vocational graduates’ detailed exam scores with administrative career records and deploy machine-learning models to identify which subject-specific grades predict labour-market success and whether standardized tests add information beyond school marks.

Conclusions

The common thread uniting my research is a commitment to causal identification of how automation and AI influence behaviour on both sides of the labour market. Three peer-reviewed papers already provide causal evidence on how automation-prone routine employment has evolved over three decades, how workers price automation risk, and how generative AI affects the supply of apprentices.

Going forward, I will extend this lens in three directions: first, by finalising the firm-side vignette study to quantify how automation expectations reshape training investments; second, by completing two large survey studies that trace how teenagers and children integrate AI into career planning and daily learning; and third, by using machine learning to evaluate vocational students’ exam grades and their labor market outcomes.

Grounded in rigorous evidence and designed for real-world application, my research aims to help societies navigate the AI transition while protecting the engine of long-run prosperity: human capital.