The Promotion

On the work I no longer do, and what it was doing for me

---

There is a moment most mornings now when four or five things are moving at once and my hands are flat on the desk.

Something is reading a document I never opened. Something else is checking a claim I made yesterday and will come back shortly to tell me it does not hold. A third is turning a thought I dictated into a draft I will either keep or delete before the coffee has gone cold. I have not typed a sentence, I have not opened a folder in months, and by eleven I am more tired than I used to be at six.

The feeling is not overwhelm, and it is not the smug pleasure of leverage either. It is closer to the tiredness of a day spent entirely in meetings, where nothing was built by you and everything depended on what you decided. Except there were no meetings, and there is nobody else in the room.

This essay looks like it is about AI. It is not, or not mainly. It is about what happens to a person who gets promoted out of the work that made them good at it, and about the particular trouble that shows up two or three years later, quietly, when nobody is watching.

Everything that follows is a single movement. A promotion nobody applied for, arriving all at once, and the slow discovery of what it cost.

---

Nobody was asked

The conversation we keep having about AI is a conversation about subtraction. What will disappear, which jobs go first, which abilities weaken, what humans still do better and for how long. These are reasonable questions and I have asked all of them, and they simply do not describe what happened to me.

Nothing was taken from me. Something was handed to me.

Sometime in the last two years, without an announcement or a conversation or a new title, I stopped being the person who does the work and became the person who directs it. One morning I was writing the analysis. Some months later I was reviewing four of them before breakfast, deciding which was right, which was merely fluent, and what the four of them together were actually saying.

That is a promotion, and it has every feature of one. The scope gets wider and the contact gets thinner. You stop being measured by what you produce and start being measured by what you allow through. You spend the day inside other people's work, except the other people are systems, and they have no history with you unless you build one, no memory of what was decided in March unless you give them one, and no opinion whatsoever about whether any of this deserves to exist.

And the work leaves almost no trace. An analysis I reject never appears in the finished piece. Neither does the contradiction I caught between two perfectly reasonable recommendations, or the version that was excellent and pointed the wrong way. By evening the output looks effortless, because everything that took the effort has been removed from it.

Intelligence is present. Organisation is not.

A room full of capable strangers

A collection of skills was never an organisation. There also has to be continuity, memory, language, boundaries, and some shared sense of what matters when two good answers point in opposite directions. All of that used to live inside the people, which is why nobody ever had to name it.

Now I have to lend the room those things every morning. A record of what we decided and why, a voice we are trying to preserve, some understanding of what may enter and what stays outside. I can write much of it down and build it into the way the systems work. Writing it down does not settle when an old assumption has to change, or which commitment wins when keeping one begins to damage the other.

Without that, I have a room full of remarkably capable strangers. Each one is excellent at the task in front of it and blind to the task beside it. The contract can be sound, the strategy intelligent, the presentation genuinely beautiful, and the three of them together can describe three different companies, confidently, in the same week.

I can ask a system to hunt for those contradictions too, and it will find some of them. What it cannot decide is which company we intend to be.

I am not producing the parts anymore. I am the only one holding the relationship between them.

That is management. It may always have been management. The good ones understood that the job was never to produce every answer but to keep competent people from producing incompatible ones, and the strange thing about this decade is that the job has been handed to almost everybody at once, including people who never wanted to manage anything, and including me.

What improvement is allowed to remove

The difficult calls are not always about correctness. Some of the work I turn down is correct, elegant and complete, and wrong because it does not sound like us.

I used to recognise people in their emails before I reached the signature. A certain way of starting too far back, an impatience in the punctuation, one word somebody leaned on too hard, the unnecessary sentence a person always seemed to need before arriving at the point. Those details carried almost none of the information and almost all of the person.

Now every sentence can be improved before it reaches me, and the question is no longer whether it can be made clearer, because of course it can. The question is what the improvement is allowed to remove. A model can follow a style guide, imitate a rhythm, even tell me when a paragraph has gone generic. What it cannot decide is which imperfection is identity and which is simply bad writing.

And some of what I used to call awkwardness was never awkwardness. It was hesitation, and it belonged there, because I am not as certain as the improved sentence makes me sound. So the decision underneath all of this is how much distance I am willing to allow between the confidence of the language and the confidence I can honestly claim.

A company can sound more professional every month and gradually stop sounding like itself. Nothing announces the moment. The sentences will simply become excellent.

Stored capital

Every promotion contains the same quiet exchange. You gain the whole and you lose contact.

Ask anyone who has run engineers. The best one on the team gets promoted, and for the first year she is exceptional, because her hands still remember the work and she can look at a solution and know within a minute whether it is honest. For a while she is running on stored capital. Then the years pass, and one day she is reviewing work she could no longer produce herself, and she does not know it, because the review feels exactly the way it always felt. The confidence still arrives. The vocabulary is still hers. Only the contact has gone.

Judgment does not announce its own expiry. It becomes ceremonial.

There is an old puzzle sitting underneath this. Ask anyone how well they understand a zipper and they will tell you perfectly well, because everyone understands a zipper. Ask them to explain how it actually works and the explanation stops continuing somewhere around the third sentence. Rozenblit and Keil called it the illusion of explanatory depth: familiarity had been standing in for understanding, and only producing the thing revealed the difference.

You know the feeling if you have been working the way I have. Something explains itself to you clearly, you follow every step, you close the window with the distinct sense of having understood. Following an explanation and being able to produce one are different capacities, they are indistinguishable from the inside, and nothing in the room is going to tell you which one just happened.

A few weeks ago I said there is almost no subject I would refuse to enter now. I meant it as a description of growth, and it is one. It is also a description of a man who removed one of the tests that used to tell him how far he had actually gone.

The work I wanted to escape

When I went looking for where my judgment came from, I found it in the parts of the job I had spent years trying to get rid of.

Not in reading. Reading gave me material. Judgment came from forming a view, acting on it, discovering that the world disagreed, and carrying the consequence long enough for something inside me to adjust, usually by a millimetre. Do that for fifteen years inside one territory and you end up with an instrument that detects something wrong before you can say what it is detecting.

I used to spend hours inside my own folders, sorting and renaming and throwing things out, moving a document and then moving it again because the first place had been wrong. I called it administration. It was inventory. Every pass told me what I possessed, what I had abandoned, what I had believed two years earlier and quietly never gone back to fix. I have not done it properly in a long time and I have not missed it once, and that is the part that unsettles me. Not the loss. The absence of the loss.

I used to type, which was slow, which was the whole complaint, and which was also slow enough to think inside. The distance between having a thought and finishing the sentence was a quiet edit running the entire time, and I never counted it as work because it never felt like work. Now I speak and the sentence appears faster than the edit used to happen, and I do not know what is no longer being caught.

And I used to spend weeks not understanding things. That is gone too. Not the learning. The humiliation. Entering a field once cost a specific and unpleasant stretch of feeling stupid with no visible end date, in which you first learned the vocabulary, then discovered that the vocabulary had solved almost nothing, and eventually, if you stayed long enough, began to understand why your first questions had been the wrong ones. I would have removed that period gladly if anyone had offered.

Somebody offered.

The old work had two jobs. It produced the thing, and it maintained the person producing it. I got rid of the first and assumed the second would carry on by itself.

Directing work builds judgment too, or no manager would ever have had any. What goes missing is smaller and much harder to see: the step I could not write around, the old file that contradicted what I was sure I had always believed, the question I had to sit with because I could not yet make a convincing sentence out of it. Those were the moments that told me where I actually stood, and I can now reach a finished piece of work without passing through a single one of them.

What thirty years actually buy

There is a comfortable version of this essay in which human experience stays sacred and machines never touch real mastery, and I do not believe it. Much of what a master learned over a career has been put into words, in books and lectures and papers and interviews, and that is exactly where these systems found it. It would be dishonest to call that access shallow simply because it arrived quickly.

Nor does time produce anything on its own. Thirty years can deepen a judgment or just make someone more attached to it. The part worth defending is narrower: learning where a good idea stops being safe to rely on.

That map is not made of information. It is made of collisions. A decision failed. A client left. A number refused to behave. A sentence caused a reaction that could not be taken back. And the map changed.

A machine can describe the edge and even warn me about a mistake I have not made. The information can become mine. The event still happened to someone else.

And now the part I would rather leave out. Every field I walk into on a Tuesday afternoon is somebody's life. I arrive briefed, articulate, able to hold a conversation that would have been closed to me five years ago, and I paid for almost none of it. I am not standing outside this defending twenty years against newcomers. I am the newcomer. There are rooms I spent twenty years building that other people now enter exactly the same way, sounding roughly as competent as I do on my Tuesday, and I cannot make the speed of their arrival the argument against them without making it the argument against myself.

The only difference I can point to is that I got the slow years first, and the collisions came with them, and I had no idea at the time that they were the part I was being paid in. Somebody starting this week will be better than I was at twenty-five and will reach the same finished work without any of it. I do not know what replaces that, for them or for me from here on.

That is not a contradiction I can resolve. It is the chair I am writing from.

The one who stays in the room

I should say plainly how this was made. I argue with a machine almost every day. It reads what I write, disagrees, holds its position, finds the counterargument I was avoiding, and does it at two in the morning without getting tired or offended. Nothing like it has existed before, it has shown me connections I would not have found alone, and pretending otherwise in order to sound cautious would be its own dishonesty.

It also risks nothing.

When a colleague tells you that you are wrong, they spend something. They put a small part of the relationship at stake, they may be remembered as the person who said the awkward thing, they have to see you the next morning. The cost does not make them right. It makes the disagreement real.

So a system can help me reach a decision. It cannot turn that decision into experience. It can make a weak idea coherent enough that I briefly forget it was weak, and it can reveal a connection I would never have found alone, and both arrive wearing exactly the same fluency. The difference shows up later, inside a contract, in front of a client, across a table, in the silence after a sentence that sounded better before it was spoken aloud. I can bring the outcome back afterwards and take it apart properly, and that is worth something.

But the systems have no afterlife. I do.

Where I stop

Go back to the morning. Four or five windows, hands flat on the desk, the work moving without me.

I am not giving any of it back, and saying otherwise would be another kind of dishonesty. What I am no longer willing to assume is that because the work keeps getting better, the person directing it must be getting better with it. Those can happen together. They are not the same process.

So I go back down on purpose. I make claims specific enough to fail and put them in front of people who have lived in the territory longer than I have. Sometimes I close the windows and build the explanation from an empty page, not because unaided work is purer, but because I want to find the point where the thought stops continuing. None of it is efficient. Efficiency is what removed it.

The systems will keep going, because continuation is their natural state, and they will produce another version, another argument, another eighty pages, and they will do it well. Management begins where continuation stops being useful. Nothing on the screen will tell me when the work has become enough, when a collection of brilliant parts has stopped forming a whole, or when the language has become better while the identity has become thinner.

And the work can now look finished long before I am ready to put my name beneath it. The old process kept those two moments close together, because producing the thing took long enough that I arrived at both at once. Now the distance between them is mine to notice, and nobody else is going to measure it.

You have the same promotion. Nobody asked you either.

So the question is not how much further you can reach, because the answer is a great deal, and it will keep being more. It is whether you still recognise the point beyond which you should not put your name.

That edge was never the limit of what you could do. It was the outline of who was doing it.