["Journal of Organizational Behavior, EarlyView. ", "\nABSTRACT\nAs organizations increasingly adopt human‐AI teams (HATs), understanding how to enhance team performance is paramount. A crucially underexplored area for supporting HATs is training, particularly helping human teammates to work with these inorganic counterparts. Indeed, research on training for HATs is limited, often relying on human team training frameworks, failing to consider the humans' expertise‐based training needs and how this may affect collaboration with AI. To bridge this gap, we interviewed competitive eSports athletes (N = 22), a group experienced in training with AI, to discuss the gaps in current human–AI training and their desires for future training that better supports humans in AI‐integrated work. Using the quantitative ethnography (QE) tool, epistemic network analysis (ENA), we examine these training needs and how they vary based on the participants' task expertise. Our findings indicate that current training methods focus on using AI for taskwork training, with significant expertise differences identified due to diverging perceptions on this taskwork focus as well as tensions related to balancing adaptability with predictability and clashing attitudes toward training with AI. Future training must evolve to deepen understanding and trust between humans and AI, focusing on socio‐emotional bonds and role awareness to offer greater benefits for teaming. We conclude with three actionable recommendations for organizational research on training for HATs to expand these findings to broader contexts.\n"]