Everyone is a middle manager with AI

Everyone is now a middle manager with AI. This is the simple reason why everyone seems exhausted with AI.

When you are an individual contributor (without AI) and have only a manager above you, your limited job is to create output, not check it as much. But if you have managed someone, you know how exhausting it is.

To be a manager is to get leverage out of your time. Now, say you want to get leverage out of AI. Say you work at a VC fund and are evaluating a fashion marketplace. You ask AI to map recent trends, estimate the market size, simulate customer behaviour and mock up what the product could look like. It gives you an investment memo, a model and a surprisingly polished set of screens. The output arrives quickly. But each piece contains dozens of decisions and assumptions. You cannot check every source, formula, conclusion and micro-interaction in the UI, yet you are responsible for the recommendation. So you make dipstick evaluations and review the work with squinted eyes, dilated pupils and an angry concentration face. After 5 hours and hundreds of decisions, you are exhausted. The work wasn’t fully yours, but the responsibility was.

Let’s also speak about the leverage ratio. To get leverage out of your time, you need to get more than you invested. The moment that ratio gets closer to 1.2 (20% more leverage than if you were to do it alone), you start feeling like the resource isn’t as useful. Your direct report feels like a star when they give you 1.5x leverage. That’s a high bar. Some models and tools have been able to clear this bar. But not most of them. And not for all use cases. For sure, this has many people feeling like AI has been thrust upon them without the leverage. (For the record, I am not one of them. I am still largely an AI optimist.)

In addition, if you have high standards as a manager—e.g. no errors in Excel, smashing UI designs, 2nd-order thinking in documents, deeply thought-through architecture, very empathetic emails or edgy marketing copy—then you have a bloody hard time explaining to LLMs how to bring that to life. Sure, skills and agents can help and get you 70% there, but that last 30% is very painful. And you know that’s where the value lies. There are 2 potentially frustrating scenarios here: either the output is almost there and you spend an enormous amount of time perfecting it, so it doesn’t save you much time; or you never reach a point where you are fully satisfied because the AI just didn’t have the right context or the right water supply to create something just right at that instance.

The silver lining: I believe that, like good people management, this skill can be learnt. And perhaps that’s how you can promote yourself to upper management!