Bottom-Up versus Top-Down: The Dual Pathways of AI Adoption in the Workplace

Technological innovation continues to reshape the organization of work, but the rapid diffusion of artificial intelligence (AI), particularly generative AI, has sparked renewed debate among researchers, policymakers, and the public. While previous studies have highlighted the productivity-enhancing potential of AI and even a reduction in within-firm inequality, these analyses typically assume that AI adoption occurs through formal, firm-led implementation. In reality, the growing accessibility of AI tools enables many employees to integrate such technologies autonomously into their work routines. Distinguishing between formal (employer-driven) and informal (employee-initiated) AI adoption is therefore crucial for understanding who benefits from AI diffusion and how it affects inequality, training, and productivity. This paper provides the first representative evidence on the prevalence, determinants, and consequences of formal versus informal AI adoption at the workplace level. We ask: (1) How widespread is AI use in German workplaces, and how does its intensity differ across users? (2) What distinguishes formal from informal AI adoption in terms of worker and workplace characteristics? and (3) What are the implications of adoption formality for training participation, productivity, and job quality? Addressing these questions is essential for understanding whether AI adoption reinforces or mitigates skill and wage inequalities. Our analysis uses microdata from the second wave of the “Digitalisierung und Wandel der Beschäftigung” (DiWaBe~2.0) survey, conducted in 2024. The dataset links detailed survey responses from 9,800 employees with administrative labor market records, providing a unique opportunity to examine within-worker changes in AI use and job characteristics between 2019 and 2024. We estimate linear probability models controlling for a rich set of pre-determined worker, job, and firm attributes to account for potential selection into AI use. AI use is widespread but uneven: roughly one in three employees report using AI, though only about one third of these use tools introduced formally by their employer. Informal adoption, that often involves generative AI tools such as ChatGPT, is driven primarily by highly educated and digitally literate workers. Formal AI implementation, by contrast, is strongly associated with increased training opportunities, greater managerial oversight, and higher perceived productivity, even after adjusting for intensity of use. The results highlight that informal, bottom-up AI adoption plays a central role in workplace digitalization. However, because it is concentrated among already advantaged workers, it risks amplifying existing inequalities unless complemented by inclusive, firm-led strategies and targeted skill development policies.