We argue that it is beneficial and epistemologically necessary to engage virtuously with a practice and that modern AI tools might undermine such virtuous practices. We suggest that chatbots are predominantly trained to provide actionable results, potentially obscuring uncertainties and complexities. The more these systems are utilized to handle tasks for us, the more we get detached from the details of the processes. This might support a focus on understanding higher-level issues, but it could also mean a loss in—what we term—low-level understanding. This low-level understanding might not be necessary for all practices. However, we argue that there are certain domains and situations for which it is vital, especially where interventions to prevent faults, e.g., in routine operations, are required. A particularly interesting case is how this relates to moral understanding connected to the understanding of the involved practices and processes. We draw parallels to other phenomena related to the extended mind, such as the introduction of calculators reducing the emphasis on mental arithmetic, or mobile phones being navigational devices diminishing the need to develop a mental map of an area. Similarly, the advent of complex programming languages like C++ has reduced the perceived need for low-level languages like Assembler. However, proficiency in these ‘obsolete’ practices remains virtuous and beneficial, as seen in fields like cybersecurity. The narrative of the detached CEO, in contrast to the one who worked their way up the corporate hierarchy, knowing every process in the firm, would be a more general example. Moreover, we explore a relational perspective, emphasizing the importance of epistemic practices in fostering relationships based on mutual understanding and shared knowledge. Finally, we argue for a culture of appreciation for the harder, more complex path to integrate AI tools as part of a critical methodology rather than a replacement for it.