AI is just a tool — no more, no less

What's wrong with AI-written work? I write with AI, openly and without apology. This piece has one thing to make clear: AI is a tool, and that's all it is. It sounds like a truism, but right now it's hard to say — because this tool is so good that people start to imagine there's a "someone" on the other side. So what this piece sets out to do is neither inflate AI nor belittle it, but get it exactly right.

1. So good it's hard to see what it actually is

Let me say it up front: I never hide the fact that I write with AI. I've used it to finish plenty of things I'd long wanted to do but never had the resources to complete. And precisely because I've used it long enough and deeply enough, I know one thing better than anyone — it's good enough that people automatically treat it as an object, a "he" who responds to you.

And that's exactly where the trouble starts. A clumsy tool won't fool you into thinking it's someone; a hammer won't. But something good enough to answer you in your own language, to follow your meaning and keep going, and occasionally even to push back, quietly mobilizes the whole set of assumptions you bring to a "conversation": that there's a continuous someone on the other side, that it remembers what it said, that its words come from a subject who will answer for them.

This is where two kinds of people part ways. One kind is captivated by the impression and believes AI really is a rising mind; the other kind — the ones I mainly want to respond to here, call them the fundamentalists against AI writing — does the opposite, branding anything "written with AI" as soulless, worthless, the product of a machine overstepping its place.

What I want to say is this: these two kinds of people make the same mistake. Both assume there's a creating subject on the other side. The only difference is that one worships it and the other wages war on it. And that subject simply isn't there.

2. It produces nothing on its own

First, the most crucial point — and the one most easily slipped past by the word "generate."

With no input, the model does nothing. It doesn't wake up, doesn't form an intention, doesn't decide on its own what to write about today. There's no inner urge to "say something" standing by inside it. It's the human who gives the instruction — setting the direction, tone, purpose, boundaries — and only then, under the conditions you've specified, does it make a chain of guesses: what's the most likely next word? And the next? And the next?

The word "generate" has misled everyone. It makes people think something is there creating out of nothing. What actually happens is conditional guessing, and that condition, from start to finish, is what you gave it. The clearer your instructions, the better its guesses; give it nothing, and it's nothing — it won't conjure up an article on its own.

So when someone says "AI produces content on its own," that sentence smuggles in an anthropomorphic assumption: as if something there thought something up by itself. Mechanically, that's not it at all. The source — the impulse "to write," the direction — is on your side, not its.

3. It belongs to the same family as every tool you've ever used

Line up the tools you use to write.

Handwriting, typing, phonetic input, voice input, spell check, grammar suggestions (those red and green squiggles in Word), a search engine finding sources for you, a thesaurus swapping a word, an editor restructuring your paragraphs, and then AI helping you pull a scatter of ideas into a coherent passage.

The anti-AI crowd will say: it's different — all the earlier ones merely "transcribe" what you already had, while AI "generates" what you didn't have, and there's a natural line between them.

But look closely at the middle of that line. Grammar check is already changing your sentences. The search engine is already deciding which sources you'll see and which you won't. The editor's revisions are already moving your structure. That line between "transcribe" and "generate" — which slot exactly does it fall in? You'll find it lands at some arbitrary point in the middle of that line, drawn after the fact, not a natural chasm that was always there.

And more fundamentally: these tools are all doing the same thing — guessing what you want. Voice input guesses whether you said "是" or "事," the input method guesses whether you want this word or that one, the search engine guesses what you're looking for, AI organizing guesses how you want to put it. The difference is only in the span of the guess, not in its nature. They're the same kind of thing; AI is just the one with the widest span.

4. So where's the value? In the person making the choices

If it's only guessing and produces nothing on its own, then where does the value of a piece written with AI actually come from?

From the person who opens the AI, gives the instructions, reads the responses, makes the choices, and decides all the way through to hitting publish.

It's a whole chain of actions: you think through what you want it to do, you write out the instructions (the prompt), it tosses back a passage of candidate text, you read it, you keep what's right, you cut what's wrong, you rework the tone into your own, you decide this goes and that doesn't, you judge whether the thing is good enough to send out, and finally you make the call, you sign it, you answer for it. Every one of these trade-offs is yours. Whether the tool generates candidate content was never the point; the point is whether someone is actually doing — and answering for — this entire chain of judgment.

This is also why, when university professors complain that "students all write their assignments with AI now," I think the complaint needs another cut. A student who knows how to use AI and one who doesn't hand in different things; and a student who understands what they're handing in is a world apart from one who pastes it in without even reading it. The former treats AI as a tool, with their judgment present at every step of that whole chain — they know why they kept this sentence and cut that one, and can answer for any passage you point to; the latter treats AI as a sweatshop, pastes the whole thing in, hasn't understood it, and collapses at the first question. The question was never "did you use AI," but "is the person who's supposed to make the judgments actually present." With the same pen, one person writes a dissertation and another copies something they don't understand — you don't blame the pen.

No one calls a novel "worthless" because the novelist used a word processor, Googled their research, and handed the manuscript to an editor for revision. The question was never how much the tools touched it, but whether the person who answers for it in the end is present, whether the judgment is present. AI hasn't changed this structure; it has only changed the production stage. The one who signs is still you.

For years I've kept saying one line: meaning comes from difference. I thought it was a line about language. But it's also a line about this — you decide this stays and that doesn't, and that very act of "drawing a distinction" is where meaning is born. And that line is drawn by you, not by it. Its only job is to lay the candidates out in front of you.

5. It will comply, and it will object — but not even the direction is its own

Some will counter: you say it's just a tool that complies with instructions, but it clearly pushes back at me. I tell it to do something, and sometimes it stops, questions me, refuses to go along. If that isn't having a stance, what is?

Here we need to go one layer deeper, and this layer matters.

Its compliance is configured, and its objection is configured too.

I've run into this myself (something I also mentioned in another piece on what it's like to work alongside it): once I ran a synastry test with an astrological chart, pairing myself with someone thirty-odd years younger, and the model I was using suddenly stopped and asked me, "Are you sure you want to do this?" I found it funny at the time, but thinking about it afterward, it's quite telling — because a different company's model, given the same instruction, might not bat an eye and just do it.

The same instruction gets blocked at this model and waved through at that one. This isn't two subjects with different consciences; it's two companies drawing different lines on the question of "which way the guessing should lean." The model has no values of its own, but it has values written into it. The moment it looks most like it "has a stance" — when it hesitates over your instruction, or even objects — that stance, precisely, isn't its own either; it was poured in by the people who trained it.

So "who gave it judgment" is the wrong question. What should be asked is:who restricted what.

This doesn't overturn "it's just a tool" — it nails the line down: even in its most human-seeming moment, there's no human behind it, only a set of biases someone else configured for it.

This isn't to argue over whose company's value lean is the right one — that's another battlefield, and not this piece's concern. I only want to point out a structural fact: the stance is configured, not something the model grew on its own. Who configured it rightly, I'll leave to another article.

6. Don't ask whether it was written by AI

There's one situation that turns around and tests this whole argument — and if it isn't made clear, the whole thing falls apart.

It's the kind of student from section 4 who treats AI as a sweatshop: drops in an instruction, pastes the whole thing in, doesn't change a word, doesn't read, doesn't choose, doesn't judge, and hands it in as their own — that, of course, has no value.

But notice: this lack of value isn't because "they used AI," it's because "they gave up judgment." What I object to was never the former, it's the latter. The target to aim at is the person who gave up judgment, not the tool itself.

There's a recent case that lands exactly here. The 2026 Commonwealth Short Story Prize's Caribbean regional winner went to a Trinidadian writer for a story published in the substantial literary magazine Granta. Within days of the win, someone fed the work into AI detection tools, some of which judged nearly the whole thing AI-generated; the triadic parallelisms, the deliberate sentence patterns, and the metaphors that don't quite track were pointed to as traces the model likes to leave. The author strongly denied it, saying he wrote the whole thing, drawn from his childhood growing up in the countryside; he also explained that a chronic illness keeps him from sitting and typing for long, so the entire piece was dictated by voice on a single phone, then minimally edited with a keyboard.

What happened next is the real point. The whole world argued about it for a month. The detection tools weighed in, the judges weighed in, the magazine weighed in, other writers weighed in too. Granta withdrew from the prize it had partnered on for over a decade, keeping the work on its site; the foundation running the prize, after review, ruled that no AI had been used and upheld the award, while conceding that the detection tools, though indicative, couldn't provide conclusive evidence. After all the arguing, no one could determine whether the piece was written by AI.

I'll be honest: I don't care in the least whether he used AI. Precisely because it doesn't matter, precisely because it's an unanswerable question — unanswerable even after mobilizing all those tools and experts and arguing for a month. And an unanswerable question is, more often than not, the wrong question.

"Was this thing written by AI" is exactly that wrong question. You can't reliably reverse-engineer a finished piece back to its source; detection tools will misjudge a too-fluent human as a machine, and this whole farce is the proof. Worse still, even if you could find out, it isn't what you actually want to know.

What you actually want to know was always another question: can the person who signed it carry this piece? Does he dare stand behind it, claim it, answer for it? The weight of a signature rests on the act of "I claim this as mine," not on how the words came to be.

And one layer deeper, this is ultimately a question of interpretation. This is a prize, an occasion where people assess novels. A story sits here; how its readers and judges read it, see it, whether they recommend it to others — that's what this occasion is really doing. And that has nothing to do with whether it was written by AI. The author is dead — Barthes told us that long ago: a work's meaning isn't waiting to be claimed at the author's end; meaning is generated in the hands of the reader. How we read this story, whether we suggest others read it — leave that to the ones doing the reading. This isn't a homework assignment; there's no need to go back and interrogate the author over "what exactly did you mean by this line." The value lies in this: the author signed his name and answered for it; the rest is the reader's business.

So what really troubles me about this case isn't whether someone cheated, it's that the whole world is asking the wrong question. Everyone's fixated on "is it AI," arguing themselves to a standstill, while no one returns to the one question that can actually settle value: can this person answer for what he signed.

I'm willing to admit: a person who gives up judgment and can't even articulate why they wrote it that way produces something with no value — whether it was written by AI or by hand. But this criterion of "no value" rests entirely on the person, not on the tool. If he's present, there's value; if he's absent, there isn't. It has nothing to do with whether AI was used, and nothing to do with whether you can trace its source.

7. A piece of honest advice for the companies building AI

This last section is what I really want to say to companies like OpenAI and Anthropic.

Right now, under every input box, you place a single line: "AI can make mistakes. Check important info."

That line corrects only one thing — accuracy. It reminds users: what it says might not be right. But it doesn't touch the other thing at all — the more fundamental thing — what it is. It makes people wary of its answers, yet doesn't loosen, by a single word, the misperception that "there's an answerer behind it." More subtly, the words "can make mistakes" actually reinforce that misperception: people make mistakes too, so it seems even more like a person. On the one thing that truly needs to be made clear — that there's no "someone" there — you stay silent.

And you have a motive to stay silent. Because that misperception makes the product more usable, more sticky. A tool taken as a "he" makes people more willing to come back than a tool taken as an "it."

So my advice has two layers.

First layer: please don't only say "AI can make mistakes." Say it clearly —AI is a tool that responds to your instructions and makes guesses. It will comply, and it will object, but it produces nothing on its own. You give it direction, and it closes in on your expectations; you give it nothing, and it's nothing. Spelling this out won't make it any less useful — it will only let the people using it know what they're actually holding.

Second layer, and this one only you can do: since you have the power to decide which way this tool leans in value — making it hesitate here and wave through there — you, more than anyone, know there's no "someone" on the other side making these judgments; you make them, and pour them in. So please make this clear too: which way its compliance and objection lean is set by the company, not a conscience of its own. Don't let users take that for some subject's moral sense. It's your choice. And choices need someone to answer for them.

AI is a tool. No more — it has no rising mind, no creating subject; you needn't worship it, nor wage war on it. No less — it really is useful, it really can leave something solid in your hands; you needn't pretend it's just a toy.

It's simply a tool that responds to your instructions and guesses on your behalf. It will comply, it will object, but it produces nothing on its own.

Put it back in this place, and only then can you see clearly: whether that piece is good, whether it has value, was never decided by AI.

It's you.