In January 2023, the American tech outlet CNET quietly published 77 AI-generated articles. For the first few weeks, nobody noticed. Then, three months in, an outside review confirmed errors and plagiarism, and CNET had to issue a public apology and overhaul its AI guidelines from top to bottom.
That same year, an even stranger episode played out in an American courtroom. A lawyer cited fictitious case law invented by an LLM in his court filings and was sanctioned by the judge. The rulings he had cited simply did not exist.
These two incidents point to the same truth. Generating text has become dramatically faster. Judging whether that text can be trusted has not. If anything, it has become harder.
Editors stand at the center of this shift. On one side are predictions that book editors will disappear; on the other, arguments that they will matter more than ever. It is worth working out which side is right.
The Faster the Tools, the Higher the Cost of Judgment
Let's start with the most intuitive misconception. The expectation is that if AI can generate drafts quickly, editors will have less to review. In practice, the opposite is happening: as generation speeds up, the number of manuscripts that need handling grows right along with it.
When producing a single manuscript took a week, there was a hard limit on how many an editor could handle in a month. Now AI turns out several drafts a day — and that many more land on the editor's desk.
What's more, sifting each manuscript for factual errors and broken context has become trickier. An awkward sentence written by a human is its own warning sign; smoothly polished AI text offers no such clue. The errors are buried deeper.
In other words, as the cost of generating text approaches zero, the unit cost of editorial judgment goes up. Making things has become easy; deciding whether what's been made can be trusted has become harder.
So What, Exactly, Does an Editor Do?
If you define the editor's job as catching typos and smoothing awkward sentences, then perhaps AI can take it over. Spell-checking software has been a fixture of editorial offices since the early 2010s, and today's AI will only do that work better.
But an editor's real competence lies one step earlier. "Do readers need this manuscript?" "Is there a reason to publish it now, at this moment?" "What effect will this book have, and on which readers?" Asking these questions is the editor's actual job.
An LLM cannot frame questions like these. It only answers the prompts it is given. Asking what book ought to be made — weighing the market, the timing, and the reader all at once — remains human territory. And that territory is the most essential part of what publishing is.
Mediating Between Author and Reader
Editing is the work of mediating between an author's intent and a reader's understanding, moving back and forth between the two. Watch how a single book actually gets made and this becomes concrete.
Authors write what they want to write. Readers read what they want to read. The two do not align automatically. The passage in chapter five that the author considers most important may be the part the reader finds most tedious. Where the author's obsession collides with the reader's curiosity, the editor serves as the connector.
The tension and compromise that arise as editor and author refine a manuscript together are what have given books their grain. But AI-generated text carries no authorial intent in need of that mediation. No voice, no fear, no traces of abandoned drafts. As AI-assisted manuscripts multiply, the new editorial challenge becomes how to bring out the real author's voice — how to preserve the human fingerprints beneath AI's polish.
Books Don't Get Recalled
Some readers will have gotten this far and asked: "Won't editors end up automated too?" Tech optimists do make exactly that argument. Once LLMs handle style feedback, reader-response prediction, even title optimization, the editor's functional role can only narrow. Some digital media companies have already cut editorial staff in favor of AI review pipelines — and since those companies haven't collapsed, that stands as evidence the approach can work.
It is hard to declare that direction flatly wrong. But one distinction has to be made: is what those companies produce a book, or text built to harvest traffic?
If traffic is the goal, an AI pipeline is efficient. When some of the content turns out to be wrong, you take it down quickly and move on to the next piece. The losses get spread out.
Books are different. Once a book has been published and distributed, it cannot be recalled. You can't round up and destroy every copy readers have bought. An error, once printed, is permanent. That asymmetry is the essence of the book as a medium — and it is what sets the editor's role apart from every other text industry.
Here, the editor becomes the last line of defense. And the job is to doubt. Asking "is this actually right?" is the core professional skill. AI does not doubt the text it produces, so doing the doubting is a human's work.
The CNET debacle and that lawyer's courtroom sanction happened in the same place: the spot where doubt belonged and none was applied. When a book editor makes the same mistake, the consequence is not one person's sanction — it is an entire book. The scale of the impact is different.
The Editor as the One Who Judges
So if I had to sum up in a single sentence what an editor does, I would say: an editor is someone who judges.
Judgment is not merely processing information — it is taking responsibility. The decision to send a manuscript out into the world puts the publisher's name and the editor's eye on the line together. AI can recommend, analyze, and draft. But the responsibility for the decision to publish ultimately rests with a person. AI cannot carry the weight of that decision on anyone's behalf.
That responsibility is not going away. If anything, it grows heavier as the tools grow more sophisticated. Deciding when to use the tools and when to stop them — how far to delegate to AI and at what point human judgment takes over — becomes the editor's new job.
What This Means for Korean Publishing
The implications of this shift for Korean publishing are clear.
An editor's eye becomes the publisher's point of differentiation. Less time goes to fixing typos and awkward sentences; more goes to fact-checking and judgment. Editors handle more manuscripts in the same hours, but the doubt applied to each runs deeper. In an era when anyone can produce the average-grade content AI can produce, the eye that discerns — above that average — what is meaningful and what fits the moment becomes the publisher's asset.
AI guidelines become a publisher's identity. Publishers that draw a clear line — AI up to this stage, human judgment from that point on — will earn trust. Those with blurry boundaries will take a far harder hit when something goes wrong.
The smaller the publisher, the more judgment matters. A large house can filter out one person's mistake through multiple layers of review. A small press or a one-person publishing house — a common format in Korea — has no such safety net. A single editor's judgment decides the fate of the entire book. For one-person publishers to survive the AI era, accuracy of judgment becomes their greatest competitive edge.
The Real Editor Is the One Who Stops
Publishing's timetable has not sped up because of AI. A first draft now arrives in eight minutes, but the review that turns that manuscript into a book still takes three months. What got faster and what stayed the same have split cleanly apart.
The tools will keep getting more refined, and text generation will keep getting faster. The real editor is the one who deliberately stops inside that speed. The one who pauses once more in front of AI-generated text, who asks "is this actually right?", who takes responsibility for the book that carries their name.
We have entered an era in which the judgment to stop the tools matters more than the skill to use them well. And that position is not disappearing — it is coming into sharper focus.
