The Last Human Job is Taste

Maxwell Perkins, an editor at Scribner's, once received a manuscript of more than a thousand pages from an unknown writer. The year was 1929. It was shapeless, overgrown, and impossible to publish as it stood. Perkins spent months with Thomas Wolfe cutting and restructuring it into Look Homeward, Angel, the novel that made Wolfe's name. Perkins also discovered Fitzgerald and Hemingway, and he insisted to the end that an editor creates nothing and only releases the energy already in the work. He was wrong about the nothing. What he did was the rarest creative act there is. He knew what to keep.

I have been thinking about Perkins as generative tools collapse the cost of making things to almost zero. For most of history, the bottleneck in creative work was production. Making the thing was slow and expensive, so the people who could make things held the value. That era is ending. A model can now generate a thousand logos, a hundred drafts, or a whole campaign before lunch. When making becomes infinite and free, the bottleneck moves. It moves to judgment.

From Making to Choosing

The scarce act is no longer generation. It is selection. When everything is possible and most of it is mediocre, the valuable person is the one who can look at a thousand options and know which one is alive. That is not a mechanical skill. It is taste, and taste is stubbornly human.

The radio producer Ira Glass once described the engine of a creative life as a gap, the distance between your taste and your ability. For beginners, taste runs ahead of skill, and the disappointment of that gap is what makes people either quit or improve. Generative tools have quietly inverted the problem. They hand everyone enormous ability overnight. What they cannot hand anyone is the taste to use it well. The gap has not closed. It has only moved to the one place a machine cannot follow.

Taste Is Earned, Not Issued

Here is the part the technology cannot shortcut. Taste is not a setting. It is the compound interest of experience, the residue of thousands of small judgments made and lived with over years. You learn what is good by making things that were not, by sitting with the disappointment, by watching an idea you loved fail in front of an audience and an idea you doubted quietly work. Every one of those moments deposits something. Eventually the deposits become an instinct you can no longer fully explain, only trust.

That is why taste arrives late and unevenly, and why it cannot be downloaded. A model has seen everything and lived nothing. It can reproduce the surface of good work because it has ingested an ocean of it, but it has never made a bad decision and felt the cost, never defended a choice in a room, never carried a project long enough to learn what it was really about. Experience is not just exposure. It is exposure plus consequence. The machine has the first and will never have the second.

This is also the quiet answer to the anxiety of the young creatives staring at the gap. The gap is not a flaw in you. It is the apprenticeship. The discomfort of knowing more than you can yet execute is the exact mechanism by which taste is built, and it is the one advantage no amount of compute can hand to someone who has not done the years.

Editing Is the Work Now

If production is solved, then editing becomes the job. Not editing as proofreading, but editing as the act of deciding what a thing is about, what belongs in it, and what has to go so the rest can breathe. Perkins did it with a pencil and a thousand pages. A creative leader does it now with a hundred machine-generated directions and the conviction to kill ninety-nine of them.

The obvious objection is that the models will simply learn taste too, that this is only a matter of more data and another generation of training. I do not think so, and the reason is structural. A model optimizes toward the center of everything it has seen. It is an averaging machine, and it is extraordinary at it. But taste is frequently the refusal of the average. It is the willingness to prefer one strange, specific, defensible thing over the thousand competent options, and to stand behind that preference when it would be easier to regress to the mean. That requires a point of view, and a point of view is not an output. It is a position someone is willing to defend.

This is also why design and language cannot be pulled apart. To edit well, you have to know what you are trying to say. The cut is a sentence and the layout is an argument, and both are expressions of the same conviction. The tools will keep getting better at producing options. They will not get better at wanting one of them more than the others.

The Comfort of the Curator

There is a quiet relief buried in all of this. For years the fear has been that the machines would take the creative work. What they are actually taking is the manufacturing of it, the part that was always the most tiring and the least human. What they leave behind is the part that was always the real job. Deciding what is worth making. Knowing what to keep. Standing behind a choice when the easy thing would be to generate ten more.

The future does not belong to the fastest producer of options. We will all have that easily accessible. It belongs to the person with the judgment to choose, the taste to recognize the one good thing, and the experience, hard-won and unrepeatable, to know the difference.