And we can train AI agents this way, beyond maximizing a score or just following some thin rules. This is called relational alignment. We align with the process of care.