I asked Claude to evaluate a leadership conflict and got back safety-first, therapeutic feedback that was completely useless for my situation. When I told it to assess me through the lens of leaders I actually respect (Aurelius, Jocko, Chanakya, Musashi), the feedback became sharper and genuinely useful. AI models default to one evaluative framework. They won’t switch unless you tell them to. Don’t accept the first frame. Set the standard yourself.

---

I had a sharp exchange with someone I’m close to. The kind of relationship with a significant age difference, where one person has invested heavily in the other for a long time. It was a real conflict about leadership, duty of care, and how you handle correction when someone under your authority won’t receive it.

I brought the exchange to Claude and asked for an honest assessment. The feedback came back fast, and it was all safety-first language about de-escalation and prioritizing the other person’s emotional experience over the standard I was trying to hold. While it was articulate and it sounded wise, it was completely useless to me, because it was evaluating my situation through a framework I don’t operate in for this particular relationship.

The model defaulted to a therapeutic lens: de-escalate, validate, prioritize emotional safety above all else. That’s one school of leadership, but not mine. So I asked Claude to assess the same situation through the lens of leaders I actually study and respect: Marcus Aurelius, Jocko Willink, Chanakya, Musashi. Classical command traditions where authority is earned, standards are non-negotiable, and consequences are how people develop.

The result was night and day. Instead of telling me firmness was the problem, Claude challenged how I was being firm. Was I losing composure in the moment? Was I threatening consequences I wasn’t prepared to follow through on? Was I repeating corrections I should have enforced years ago? The critique was sharper, more specific, and harder to dismiss. It stopped telling me to soften and started asking whether my firmness was disciplined. Those questions held me to a standard I actually respect.

Same situation. Same facts. But completely different feedback.

These models have a default evaluative posture that leans toward de-escalation, psychological safety, and consensus-building. Useful in many contexts, but if you lead from a tradition built on accountability, direct authority, and earned hierarchy, the default will misread your leadership as dysfunction. And the model won’t switch frames unless you tell it to. It doesn’t ask “what kind of leader are you?” before offering feedback. It assumes one model fits all.

If you’re using AI to pressure-test your decisions, don’t accept the first frame it offers. Tell it who you are, what tradition you lead from, and make it work within your values, not its defaults. The model is a mirror. But if you don’t adjust the angle, you’ll see someone else’s reflection.

…