There is a version of the AI conversation happening in boardrooms right now that goes like this: coordination work is the most automatable work we have, most of middle management is coordination work, therefore we can run this company with fewer layers. Roughly one in five organizations is reportedly using AI to flatten structure, and the roles taking the hit are the ones sitting between the people doing the work and the people setting direction.
I want to say two things about that, and they do not agree with each other.
The first is that a lot of it is fair. I have run multi-site operations. I have owned P&Ls. I have sat in the weekly meeting where six managers take turns reading status updates that a shared document would have handled. There is real coordination overhead in most companies, and some of it exists only because information moved slowly. If the information now moves instantly, defending the layer that used to move it is not leadership, it is nostalgia.
The second is that we are about to make a very expensive mistake, and it will not show up for five years.
The middle was never only coordination
Ask any senior operator where they actually learned to lead. Almost nobody says a training program. They say the first time they had six people reporting to them and one of them was struggling. The first time they had to tell a customer something the customer did not want to hear. The first time they owned a number and missed it, and had to walk into a room and explain why.
That is the middle. Not the status reports. The reps.
A first-line manager role is the only place in most companies where the stakes are real but survivable. You make a decision, you are wrong, you find out fast, and the company absorbs it. That is how judgment gets built, and judgment is the one thing that does not transfer through a document, a course, or a model.
If you delete the rung, you do not delete the need for the skill. You just move the moment of truth further up, where the mistakes cost more and the person making them has never made a small one.
Coordination and judgment are not the same job
The useful move here is not to defend or eliminate the middle. It is to stop treating it as one thing.
Take any manager role and split it down the center. On one side is coordination: scheduling, status, routing, reporting, chasing, summarizing. On the other side is judgment: who to hire, who to develop, who to move on from, what to escalate, what tradeoff to make when two priorities collide and the answer is not in the policy.
The coordination side is genuinely automatable, and I have been putting AI into that side of the work across my companies. At Riderflex, at Robo Reliance, at Kura Home, the same pattern holds. AI is very good at the parts of the job that were always administrative tax, and freeing a manager from those hours is a real gain.
The judgment side has not moved at all. Not a little. Not at all. Deciding that someone is in the wrong seat, or that a customer relationship needs a hard conversation, or that the number you are chasing this quarter is the wrong number, is still entirely a human act, and it still requires having done it before.
So the question is not how many managers you need. It is how many people you need who are building judgment, and where they are going to build it.
Manufacture the reps on purpose
If your org chart is getting flatter, and for a lot of companies it should, then development stops being something that happens by accident and becomes something you have to design.
Give people small ownership early. A line, a site, a service area, a segment, something with an actual number attached to it. It does not need to be large. It needs to be theirs, with real consequences and a short feedback loop.
Put people in front of customers and in front of problems, not just in front of dashboards. A leader who has only ever seen the business through reporting will make confident decisions about a reality they have never touched.
Let them be wrong while it is cheap. The instinct in a lean organization is to route every decision to the most experienced person, because that is efficient this quarter. It is also how you end up in five years with a leadership team that is one deep.
And be honest about the difference between tool fluency and leadership. Someone who is excellent with AI and can produce five times the output is valuable. They are not automatically ready to lead people. Those are different capabilities and we are already starting to confuse them.
The math nobody is running
Flattening the middle produces a clean, immediate, defensible number. You can put it in a deck this quarter.
The cost lands later, on a different balance sheet, in a year when a key leader retires or leaves and the internal bench is thin, and you go out and pay a premium for someone external who does not know your business. I spend a lot of my week in the recruiting side of this, and I can tell you that the companies buying leadership on the open market at a premium are almost always the ones who stopped building it internally a few years earlier.
Cut the coordination. That work should go. But be very deliberate about what you are cutting alongside it, because the training ground and the overhead were sitting in the same job description, and only one of them was actually waste.
Connect with Steve: linkedin.com/in/stevepurban