My Boss Has a Boss Who Has an AI. I’m Three Layers Away From the Real Decision.

The model didn’t take anyone’s job. It quietly hollowed out the middle, and now my pushback has nowhere to land.

Photo by Chelaxy Designs on Unsplash

My manager told me, almost in passing, that the decision about our team’s roadmap had already been made.

I asked who made it.

She paused. Then she said, with a kind of tired honesty I have not stopped thinking about: “Her boss showed her an output, and the output was the plan.”

That is when I realized something about how my company actually works in 2026, and once I saw it I could not unsee it.

I am three layers away from the real decision now. My manager reports to a director. The director reports to a VP. The VP, increasingly, reports to an AI system that produces strategic recommendations for the entire division. And by the time those recommendations have moved down through the layers and reached me, they have stopped being recommendations. They are the plan. They are what we are doing this quarter.

Nobody at any level made the call in a way I could push back on. There was no meeting. There was no human I could go talk to and say “wait, this assumption is wrong.” There was a model output, a director who took it as the starting point because pushing back on it would look unserious, and a manager who inherited it as a fact.

I want to be precise about what I am describing, because this is not “AI took my job.” It is something stranger and harder to talk about. AI did not take anyone’s job in this story. It quietly took the meaningful part of several jobs above mine, by becoming the thing those people now defer to. The VP did not lose her role. She lost the part of her role where she was the source of the decision. Now she is the channel for it.

And when the source of the decision is a system rather than a person, the entire chain of conversation underneath it starts to feel like theater.

Here is the specific thing that changed for me, and that I think is happening to a lot of people quietly: my pushback no longer has a destination.

In the old version of working life, if I disagreed with a plan, I could raise it with my manager, who could raise it with her boss, who could raise it with their boss, and somewhere up that chain there was a human who could be persuaded, or at least have to look you in the eye and say no. The chain was long, but it was made of people. Persuasion was possible, in principle, at every link.

Now I can raise it with my manager. She can raise it with her boss. Her boss can look back at the AI output and say, “well, this is what the analysis recommended.” And we have hit a wall, because the analysis is not a person. It cannot be persuaded. It cannot reconsider in light of context it did not see. It cannot be embarrassed in front of its peers. It produced an output, and the humans above me have agreed, individually and quietly, that disagreeing with the output is more career-risky than implementing it.

That is the part that genuinely unsettles me. The AI is not what made this happen. What made this happen is that the humans in the chain have started to use the AI as cover. “The model recommended it” is now the most defensible thing a mid-level executive can say in a room. It is, structurally, an enormous gift to whoever is afraid to make a call. And so the call gets outsourced upward to a system that cannot be argued with, and the humans become, charitably, translators of its outputs.

I do not think any individual person in my chain is doing anything wrong. They are responding to incentives. Pushing back on an AI recommendation requires you to put your own judgment on the line, on the record, in writing, in a way that nobody in middle management is rewarded for in 2026. So they don’t. So I don’t have a destination for my pushback. So the decision feels inevitable from below, even though no human actually decided it from above.

There is no clean fix for this from where I sit. I can’t change the org chart. I can’t unplug the model. But here is what I have started doing, because doing nothing felt worse than doing something imperfect.

I started writing down my disagreements. Not as complaints. As records. When a plan comes down that I think is wrong, I write a short memo to myself, dated, with the specific reason I disagree and what I think will happen instead. Half the time I am wrong. The other half, I have a paper trail that proves my judgment was good — not for anyone else’s benefit, but for mine, so that I do not slowly lose the ability to tell when a decision is bad.

I also started picking my battles more carefully. There is no point pushing back on every model-shaped decision. There is enormous point in pushing back on the one or two each quarter that are genuinely going to cost the company real money, because those are the moments when the chain briefly remembers that humans are supposed to be in it. Sometimes I have moved a decision. Mostly I have not. Either way, I have stayed in the practice of forming a real opinion, which is a muscle that atrophies very fast when nobody is asking for it.

The thing nobody is talking about, in all the AI-and-jobs coverage, is this middle layer. The robots are not coming for the bottom. They are quietly hollowing out the middle, where the decisions used to live. And the bottom — me, you, the people who actually do the work — is left with a chain of command that no longer leads to anyone who can be reasoned with.

So I will leave you with the question I have started asking myself, when a plan lands on my desk that I disagree with:

When the decision that affects you comes from a model’s output, who do you actually go talk to?

If the honest answer is “nobody, really,” then you are not working in the company you think you are working in. You are working in a translation layer. And the first step to protecting yourself is to know that is where you are standing.