The problem with adding more
Most leadership development is additive. New framework on top of old framework. New mental model layered over the existing one. The assumption is that good new thinking will gradually displace less useful old thinking. In practice, that's not what happens. What happens is that people end up with two competing ways of understanding a situation, and they reach for whichever one is more familiar under pressure. The old model usually wins. This is why so many leaders can describe the new approach perfectly in a workshop and then not apply it at all when things get difficult.
Four things that need to go
In twenty years of working with senior leaders on transformation, I keep seeing the same four mental models cause the most damage in an AI context. The first is the idea that more data leads to better decisions. In an AI environment, data is abundant. The constraint is knowing what the data means, and having the judgment to act on an interpretation when you can't be certain. Leaders who wait for certainty before deciding will be permanently behind. The second is the belief that deep expertise protects you. In a domain that's changing as fast as AI, deep expertise in yesterday's tools can make you more confident in the wrong direction. Curious generalists often navigate this better than confident specialists. The third is the idea that leadership means executing well. The most important work a leader does in a transformation is designing the conditions in which other people can execute. That's architecture, not execution. The fourth is the assumption that consistency equals competence. Holding a position in the face of new evidence used to read as strength. In a fast-moving environment, it reads as rigidity.
Why this is harder than it sounds
You can't unlearn a mental model by learning a better one. You unlearn it by repeatedly encountering situations where the old model fails, noticing the failure, and consciously choosing a different response. That process is slow, uncomfortable, and doesn't lend itself to a curriculum. It requires real situations, honest feedback, and the tolerance to sit with not knowing. Most leaders have spent their careers being rewarded for looking certain, even when they weren't. Asking them to develop genuine comfort with ambiguity is asking them to practice something that has historically been penalized.
What actually helps
What I've seen work is deliberate exposure to situations where the old model clearly doesn't fit, paired with space to reflect on what happened. Not a case study. An actual situation, with actual stakes, followed by an actual conversation about where the reasoning went wrong. It's slow. It can't be scaled to a cohort of two hundred people in a two-day workshop. But it produces real shifts. The organizations that are developing leaders who can genuinely navigate AI transformation have built the conditions for this kind of development, even if they don't call it unlearning.