The cursor blinks. The research is open in nine tabs, the coffee has gone lukewarm, and I have read enough to write a competent summary of the thing. What I do not have, yet, is the angle. That is the part of the morning the machine cannot do for me, and on the mornings it does not arrive, I drift. Hours of reading turn into hours of nothing, and then the day is gone. On the better mornings, something clicks, and the whole pile of notes suddenly points in one direction. I cannot make that moment come on schedule. I can only sit there until it does.
This is, I think, the real texture of working in a world where a tool can write a passable first draft of almost anything in seconds. The summarizing got easy. The angle did not.
I should say up front that I am not a psychologist or a cognitive scientist, and the studies I mention here are findings from particular groups of people, not settled laws about how every mind works. This is reading and reflection, not advice.
What got cheap and what didn’t
Two things used to be expensive that no longer are. The knowledge side collapsed in price. A memo, a landing page, a pitch, a first pass at almost any document, all of it can now be generated in minutes. I use AI for some of that gathering and thinking aloud myself, and it is genuinely useful for the dull middle of a task. What has not collapsed is the imaginative jump, the part where you decide what the thing should actually be about. That asymmetry is what I notice most in my own day. The cost of producing something competent fell off a cliff. The cost of having the idea worth producing stayed roughly where it was.
There is some evidence the gap is real and not just my own flattering story. Koivisto and Grassini ran 256 people against three chatbots on a divergent-thinking task. On average the chatbots won. The best human ideas still matched or beat the machines, which tells me something in human creativity is hidden by the average. A 2026 paper in MDPI’s Information sharpens the picture. Humans and AI generated ideas at near-parity, but they diverged at the moment of choosing which idea was best, picking substantially different favorites from the same shared pool. Human choices tracked context and lived experience. The machine’s tracked statistical pattern. We generate similarly. We pick differently.
This is in no way saying I am one of the best, of course.
Where the human part actually lives
If generation is the cheap part, then the value has to live somewhere downstream, and that somewhere, I am pretty sure, is judgment.
Writer Raj Nandan Sharma seems to agree; when everyone can produce something that looks decent, the advantage may shift to judgment. He goes further, arguing that “the scarce skill is not generation. It is refusal.” That is a provocative line from a personal blog rather than a measured finding, but I think it names the thing well. The work is no longer making more. It is choosing what to keep and what to throw out, and being able to say why.
Ted Chiang made a related point about writing itself in his 2023 New Yorker essay on what large language models actually do. He argues that “Your first draft isn’t an unoriginal idea expressed clearly; it’s an original idea expressed poorly, and it is accompanied by your amorphous dissatisfaction.” That dissatisfaction is the whole game. The felt sense that the thing is not right yet, the thing that makes you delete a paragraph you cannot logically defend deleting. His other line is the one I keep returning to: “Sometimes it’s only in the process of writing that you discover your original ideas.” If that is true, then handing the writing to a machine hands away the discovery, not just the typing.
Feeling human is a practice, not a stance
I want to be honest about the limits of this argument. It is comforting to say judgment is ours and leave it there, as if feeling human were a position you hold rather than a thing you do. It is not. It is a practice, and most of it is unphotogenic. The second coffee. The walk around the kitchen with no destination. For me a lot of it is music, which is about as close as I get to adult play, the felt and entirely non-useful texture of being a person rather than a producer of outputs. The first thing I did after reading what ChatGPT could do with my own writing was put on Khruangbin and not think about any of it for a while. None of that shows up in a deliverable. All of it is the soil the angle grows in. Chiang notes that the struggle to express your thoughts does not vanish once you learn to write; “it can take place every time you start drafting a new piece.” You do not get past the struggle. You just keep choosing to sit in it.
If any of this is hitting closer to home than it is interesting, the worry that the work is being hollowed out, or that you are, talking it through with a qualified counsellor is worth more than any essay.
Not beating the machine, just not handing it the wrong things
My view is that staying human in this stretch of history is not about outracing the tool. I will lose that race on speed and probably on average quality too. It is about not quietly outsourcing the parts that were mine to begin with. The gathering can go. The first ugly pass can go, sometimes. The struggle and the dissatisfaction and the moment the angle finally arrives, those I keep, not out of pride but because they are the only part of the morning that is actually me thinking.
