Cognitive automation powered by advanced intelligent technologies enables organizations to automate more and more of their knowledge-work tasks. Though offering higher efficiency and lower costs, cognitive automation exacerbates erosion of humans’ skills and expertise on the automated tasks. Letting go of obsolete skills is necessary to reap the technology’s benefits – however, erosion of essential human expertise is problematic if workers remain accountable for tasks where they lack sufficient understanding, rendering them incapable to respond if the automation fails. Though the phenomenon is widely acknowledged, the dynamics behind such undesired skill erosion are poorly understood. Thus, taking the perspective of sociotechnical systems, we conduct a case study of an accounting firm that had experienced skill erosion over years of reliance on their software’s automated functions. We synthesize our findings using causal loop modeling based on system dynamics. The resulting dynamic model explains skill erosion via an interplay between humans’ automation reliance, complacency, and mindful conduction. It shows how increasing reliance on automation fosters complacency at both individual and organizational levels, weakening workers’ mindfulness across three work task facets (activity-awareness, competence maintenance, and output assessment), resulting in skill erosion. Such skill erosion may remain obscure, acknowledged by neither workers nor managers. We discuss these implications for theory and practice, also identifying directions for future research.