Making the Invisible Visible: Understanding the Mismatch Between Organizational Goals and Worker Experiences in AI Adoption
Citation. Lee, C. P., Lee, M. K., & Mutlu, B. (2026). Making the invisible visible: Understanding the mismatch between organizational goals and worker experiences in AI adoption. arXiv:2605.03078.
Summary. This qualitative study examines why workplace AI adoption can fail even when organizations introduce AI for efficiency, productivity, and innovation. Drawing on interviews and workplace demonstrations with 16 professionals using decision-support AI systems in healthcare, finance, and management, the authors argue that adoption problems often come from a mismatch between organizational goals and workers’ actual tasks, autonomy, communication norms, and accountability. They identify structural, task, and worker-level gaps and argue for worker-centered AI integration that preserves worker control, supports collaboration, and aligns AI communication with real work practices.
Key claims
- Organizational AI adoption fails when workers are treated as peripheral to design and deployment decisions, even though they are the actors expected to integrate AI into daily work.
- Across healthcare, finance, and management, workers described six recurring gaps: poor interoperability with collaborative work, divergent attitudes across workplace hierarchies, AI taking over tasks workers preferred to perform, insufficient worker control and involvement, skepticism about performance errors and misuse, and communication methods that did not match worker needs.
- Workers wanted AI positioned around task fit, not technical possibility alone: tedious, repetitive, lower-stakes, or analytic-support tasks were more acceptable targets than high-stakes decisions requiring expertise, contextual judgment, or valued customer/patient interaction.
- Lack of control undermined autonomy, motivation, learning, and accountability; workers wanted mechanisms to verify, adjust, simulate, and contextualize AI outputs before final decisions.
- AI adoption is a sociotechnical work-design problem: organizations need to align technology with task structure, worker values, communication practices, and collaborative workflows rather than treating implementation as a tooling event.
Evidence & limitations
- Evidence comes from approximately one-hour in-person interviews and workplace demonstrations with 16 U.S. participants: healthcare workers (6), finance workers (5), and management workers (5), all with at least three months’ experience using decision-support AI systems.
- The analysis used grounded-theory coding of interview transcripts and workplace demonstrations, so the contribution is exploratory and qualitative rather than causal or generalizable.
- The sample is small and limited to three domains and decision-support systems; findings may not transfer to all forms of workplace AI, especially generative AI tools, education, government, or other organizational settings.
- The paper is an arXiv preprint; treat it as public but not necessarily peer-reviewed.