The Author Was Absent

The draft arrived via email a few weeks ago. It was polished and structurally sound. I read it the way I read most things now, which is to say I was scanning for the place where the person actually came through. The line that could only have been written by them. Where the angle was theirs and not the AI model’s.

I could not find it.

I read the draft twice. Then a third time, slowly, looking for the seam where the human had argued with the machine. There was no seam. The piece was uniformly competent and uniformly unfamiliar, the way a hotel room is competent and unfamiliar. I knew they had used AI to write it, and I had assumed there would be something underneath, the way there always used to be something underneath. A turn of phrase, a stubborn position, a half-formed idea that resisted being smoothed out. I kept looking.

A few days later we sat down to talk through it.

I asked them about a specific section. They paused. They said they would have to look at the draft to remember what they had argued there. I asked about another part. The same pause. By the third question, I realized I was conducting an interview with someone who had not been in the room when the piece was written. They had received a thing back from a system, signed it, and sent it on. The conversation we were having was about a document that, in some real sense, neither of us had encountered before.

It felt like talking to a ghost writer who was not the writer and not the ghost. Just a person standing where the author should have been.

I have been turning that experience over for a few weeks now, and what I keep coming back to is not the embarrassment of the moment. The embarrassment is the easy part of it. What I keep coming back to is the question underneath. If this is what work looks like now, what does it mean to produce something you do not own?

The CEO of OpenAI, Sam Altman, said something in February that struck me at the time and has kept resurfacing since. Asked what would matter in a job market about to be reshaped by his own product, he said: have good taste.

That was the answer. The thing he was naming as the surviving skill, in a future he is helping to build, was the very thing his product was quietly taking from the people most likely to use it.

Earlier this month the Boston Globe ran a piece called AI is destroying good writing. The argument was familiar, but the cultural temperature around the argument was not. A year ago, the same piece would have read as a curmudgeon’s complaint. In April of 2026, it reads as recognition. People are starting to say the thing out loud.

What they are saying is not that the writing is bad. The writing is fine. The writing is, very often, technically improved over what the same person would have produced unassisted. Something underneath the writing has gone missing, and the absence is becoming legible, and nobody is sure yet what to do with it.

The MIT Media Lab ran a study in 2025 that I keep returning to, in part because the conclusion is so quiet relative to its implications. They sat people down to write essays. One group used ChatGPT, one used a search engine, one used nothing at all. They measured neural activity with EEG over four months. The group that used ChatGPT showed the weakest brain connectivity throughout, reported the lowest sense of ownership over their own essays, and, when asked to quote from work they had just submitted, could not accurately do it.

That last detail is the one I cannot stop thinking about. They could not quote themselves. The work had passed through them, briefly, the way a phone number does when you read it once and dial it.

The researchers called the pattern cognitive debt. The phrase is doing a lot of work. It implies that the productivity gain in the moment is real, the same way a credit card purchase is real, but that something is being borrowed against in the process, and the bill arrives later in a form most of us are not yet equipped to read.

A separate study at Microsoft Research, also from 2025, surveyed 319 knowledge workers about how their thinking had changed since adopting generative AI. The finding that stayed with me was that the higher their confidence in the AI, the less critical thinking they reported doing. The shape of cognitive work was shifting from generating ideas to verifying ones the system had produced. The intellectual labor was still happening. It was just happening in a different posture.

A third study, on skill decay in performers assisted by automated systems, used a phrase that seems to apply much more broadly than its original context: skills atrophy without performers’ awareness. The performer keeps performing. The dashboard looks fine, but underneath, something is thinning out, and the person who has thinned is the last one to notice.

I have been trying to find the exact name for what was missing in that draft I could not place.

What was missing was harder to name than I expected. The piece was intelligent, so intelligence was not the issue. The person had clearly spent time on it, in the way one spends time choosing a prompt and editing a result, so effort was not the issue either. Voice came closer, though even voice was not quite the right word for it. What was missing was the thing that makes a piece of work feel inhabited. The sense that the person who signed the bottom of the page had taken the trip the page describes.

Cognitive ownership lives in the process of making the thing, in the wrestle that happens before the piece is finished. It is what you accumulate when you struggle with a sentence, choose the wrong word, change your mind, and end up somewhere you did not expect to be. It is the residue of having been there. You can produce a polished piece of work without it, and you can produce a flawed piece of work that is saturated with it, and the difference between the two is invisible until someone asks you a question and you have to answer from inside the thinking.

By every formal measure, name on the document, accountability for the result, the rights of authorship, the work I had read was theirs. What was missing was the kind of ownership that lets you walk into a conversation and speak from inside the piece, instead of standing next to it and squinting.

The author who was not in the room when the work was written cannot be in the room when the work is questioned.

That is what I had been looking at, in that conversation. A piece of work whose author had stepped away during the part where authorship is built, and who could not, in good faith, return to it later.

I use AI to draft. Many of the people I respect use it. The question is not whether the tools belong in the work. The question is what happens when the use of the tools removes the part of the process where ownership is actually formed, and we keep treating the output as if the ownership is still there.

We are about to find out, at scale, what an organization looks like when most of the work being signed off on by senior people is work they do not have cognitive ownership over. Decisions made by humans who cannot quote the reasoning behind them. Documents endorsed by leaders who cannot defend the second paragraph under questioning. Strategies whose authors, when pressed, would have to look at the strategy to remember what they meant by it.

The systems being deployed inside those organizations are probabilistic. They produce output that sounds confident regardless of whether it is right, and the only meaningful check on that output is the discernment of the human who deploys it. If the human’s discernment is also thinning, in the same period and through the same mechanism, there is no check. There is only the appearance of one.

That is not a technology problem. It is a question we have not yet learned how to ask out loud, about what it means to be the person whose name is on the work.

I do not think the answer is to stop using the tools. The tools are not going anywhere, and most of what they enable is genuinely useful. The answer, if there is one, is closer to a different relationship with the wrestle. Doing the part of the work that hurts a little, on purpose, because the hurt is where the ownership is built. Being willing to write the bad sentence yourself, instead of accepting the good one the machine offered, when the bad sentence is the one your thinking actually produced.

I have started doing this in my own work, more deliberately than I used to. The draft gets read three times before I let the model touch it. An awkward turn of phrase stays in for longer than it used to. I have learned to keep the hour where I do not yet know what I think, before I ask anything to help me think it. The output is sometimes worse. The ownership is unmistakably better.

I do not know how this scales. I do not know what it looks like in an organization of fifty thousand people, all of whom are quietly outsourcing the thinking that used to make them the people their colleagues trusted. I suspect we are about to find out, and I suspect the finding will not be quiet.

What I am sitting with, this week, is the question of how we hold onto authorship in a moment that keeps offering to take it from us. I do not have a clean answer. I have a draft on my desk that I read three times before I asked anything to help me with it, and I have the memory of a conversation with a person who could not tell me what they had argued in their own document, and I have the suspicion that the second of those is what most of us are quietly drifting toward.

If you are still here, and still reading slowly, that is itself the practice.

Still writing,

Yen