The Work Used to Highlight Their Skills

In the last post, I argued that managerial judgment used to get built through small, recoverable mistakes. But AI has both raised the cost of those mistakes and shortened the time managers have to develop. The frontline manager is where that pressure concentrates. I want to stay in the manager’s chair and look at a quieter problem, one that has less to do with output and more to do with whether you can still fairly evaluate your people’s skills.

Throughout my time as a manager, a primary way you developed someone was straightforward in shape even when it was hard in practice: help them stretch their understanding of their capabilities. You handed them work a little beyond their current level and let them wrestle with it, then you reviewed the results and helped them see what went right and what needs work. This lets you (and them) see where they have been growing and where they still need to develop their skills. Ultimately, their work was a view into their judgment, and most of managing their growth ran through it.

AI has obscured the view in two ways. The first is that the work which used to build judgment is largely taken by AI now. The hard problem someone had to sit with until it finally made sense is exactly the thing the agent hands back apparently solved. So, you cannot give people the same formative reps. The first pass comes back looking finished to a junior employee, and the effort that used to do the teaching never happens.

The second is that output no longer reveals capability. Someone early in their career can now produce work that looks like a senior person made it. So, when you review that work and it holds up, you have learned that the work holds up. You have not learned what the person can do on their own. You’ve learned how well they can use AI. The signal you relied on to assess people and grow them has been cut loose from the thing it used to measure.

This is where the research I mentioned last time matters. As the Wharton work suggests, people accept confident answers more readily than we would like to believe. When someone uses AI and accepts what it gives them, the fact that it works tells you almost nothing about whether they could have caught it being wrong. A manager who reviews only the result is, more and more, reviewing the AI’s competence rather than the person’s. The thing you are trying to measure, their discernment, is the one thing the results have stopped showing you.

It’s important to note that this is not the employee doing something wrong. They are using the tools they were given in the way those tools invite, and the pull toward accepting fluent output acts on all of us, experienced people included. This is a change in the structure of the work, not a failing in the people doing it.

But it leaves the manager partly blind. Understanding an employee’s skillset and goals is the whole foundation of developing them, and when the work goes quiet as a signal, you can end up trusting someone with scope on the strength of output they did not really own, or overlooking someone whose judgment is genuinely strong but whose work now looks like everyone else’s. The instrument you have used to identify the team members who have a strong foundation and are working to grow their skillset has lost some of its resolution...right when the stakes for getting those reads right have gone up.

So, the manager’s problem is not that people use AI. It is that the old way of developing them, where you watched the work and understood the person’s skillset from it, depends on a signal AI has muddied. In the next post I will get concrete about what I see working in response, the couple of changes that seem to help managers see their people again and build discernment on purpose.