The Apex Blind Spot

A lawyer friend told me that, over time, the work her junior colleagues are producing with large language models is getting closer in quality to the standard she had developed through years of legal practice.

The sight of work each week from team members whose use of AI ranges from novice to advanced provides an ongoing point of comparison. The gap, therefore, between what someone at a given experience level is expected to produce and what they deliver is visible, and that repeated exposure gives her some indication of how much difference her accumulated experience continues to make.

Senior executives can generally find ways to observe the distance between unassisted expert work and AI-assisted output from less experienced colleagues, and can track how that gap narrows as models and usage improve.

In writing my upcoming book, The AI Gap, I was reflecting that a chief executive may be a limiting case. A chief executive has no junior counterpart producing a comparable body of chief-executive work. Work that is singular, produced under conditions that apply to that role alone and judged as a whole rather than tracked as a series, leaves the person holding that role without a spread of competence against which to read what a model can now do.

The chief executive's position is unique in its performance requirements and conditions, so the AI capability gap remains a blind spot at the apex unless the chief executive intervenes personally.

The model doesn’t need to do the whole job to disrupt the apex position. Even a subtle shift in the decision-making logic or the underlying analysis creates a vulnerability. At that point, the model is already doing work the role exists to provide; yet the executive has no vantage point to see that shift.

The chief executive can still test this. They can compare their own decision against a version drafted by the company’s AI using the same source material. Unlike the lawyer, who can isolate a junior’s specific contribution against known skill levels, the executive is flying solo. They are the only author, the only judge, and the one whose own bias dictates whether the model’s output holds water.

External advisers are the logical counterpart. They handle the prep work and analysis that informs executive decisions; they could, in theory, provide the same clarity the lawyer gets from her team. But they deliver a finished product. The executive doesn't see the “how,” what the model did versus what the adviser added. And an adviser isn't going to volunteer that their expensive insight might now be automated, especially when their fees depend on the assumption that it's their work.

That leaves the executive to build the testing series themselves. While the lawyer’s juniors generate comparative data as a by-product of their day-to-day work, the executive has to curate it from scratch. They must bank their old decisions (the raw prep, the initial stance) to run against every model update. They need a log, a persistent record of every pass, because without that history, there is no series. It is a time-consuming grind, effectively carving out a second job from the time intended for the first.