There is a small moment, mid-task, that turns out to be surprisingly hard to describe.

You are writing something. An email, a summary, a note to a colleague. And the sentence finishes itself half a beat before you do. The reply is suggested before you had decided what you thought. The draft is mostly already there, and it is fine, so you tidy it and send it. Nothing went wrong. The work got done. But if someone asked you an hour later which of those words were actually yours, you would not be able to answer cleanly.

That is the thing worth sitting with. Not a crisis. Not a productivity problem. Just a quiet difficulty saying whose thinking that was.

The transfer nobody announces

The research here is young and honest about being young, so hold all of it lightly. But it points somewhere specific.

One study looked at smart replies, the little suggested responses that sit under a message, and tested what happens to the person using them. It found the tool measurably shifts what a person writes and how they behave, without anyone deciding to hand anything over.[1] The developer did not intend it. The user did not choose it. The words simply move, a little, from you toward the system, and the move is gradual rather than total. Agency, in that study's framing, is not something you either keep or lose. It is something that redistributes quietly while you are busy getting the thing done.

Which is part of why the moment is so hard to name. There was no point where you handed over the pen.

Yours, and also not

Then there is the ownership side, which a separate study named almost exactly.

When people wrote text with AI help, many did not feel they owned or authored the result, even as they were willing to put their name on it in public and call it theirs.[2] The two came apart: the public claim of authorship, and the private sense of it. And the same work found something quieter underneath. The more a person's own input had shaped the text, the more they felt it was theirs. Ownership was not a fixed rule about what counts as writing. It tracked how much of your own hand you could still see in the result.[2]

It is one study, well designed but not yet replicated, so treat it as a lead rather than a law. But it describes the feeling with unusual precision. The reason a finished piece of work can feel not-quite-yours may not be that you cut a corner or fell short. It may simply be that there was not much of you left visible in it.

It was never really about the tool

The easy version of this story is that AI is doing something to us. The more careful version, and the one the evidence actually supports, is smaller and stranger.

A survey study of employees found that adopting AI at work did not, on its own, move wellbeing up or down. What mattered was how the work itself got reshaped underneath it: whether people kept a sense of control, whether the task still had a shape they recognised.[4] The technology was not the lever. The redesign of the work around it was.

And when practitioners in creative and knowledge work describe this in their own words, the same double note appears. In an analysis of what UX designers were saying about leaning on generative tools, the relief was real, the routine parts lifted off. Alongside it sat a worry they kept raising themselves: that the more they offloaded, the more their own judgment and craft felt like it was thinning.[3] That is people reporting a perception, not a measured drop in skill, and it should be read that way. But it is striking how consistently the perception turns up.

Whose day was it

None of this is really about hours. You are not necessarily working more. What gets harder to locate is where your contribution starts and stops inside the hours you already have.

The workday used to have edges partly because the thinking had to happen inside it, at your pace, in your words. When some of the thinking arrives pre-formed, those edges stop lining up as cleanly. That is a felt description, not a finding. Nobody has measured it, and it should not be dressed up as something anybody has. But a lot of people seem to recognise it when it is said out loud.

This is a cousin of something MoMo has written about before. In The invisible in-basket, the theme was a whole category of real work that the systems around a person were never built to see. This is a nearer-in version of the same move: not the queue you carry, but the question of whose thinking went into the thing that has your name on it. It also sits next to the standby tiredness we wrote about in "The cost of staying reachable all day," though the mechanism is different. That one was about staying half-turned toward a channel. This one lives inside the task itself.

Not a verdict on you

If a piece of work does not feel like yours, that is worth reading gently. The early research does not frame it as a failure. It frames it as information: about how much of your own input made it into the result, and about a transfer of agency that happens without ever being announced.[1][2]

So there is nothing to do here, no rule to keep. You do not have to disclose more, or lean on the tools less, or claw back every sentence you send. But the next time you finish something and cannot quite feel your hand in it, it may be worth knowing that the feeling is real, it has started to be studied, and it is not evidence that you are not a real author. It is closer to a readout of how much of you made it onto the page.