I’m Begging You: Learn How to Write With AI

Bret Stephens is right about one important thing in his New York Times column, “I’m Begging You: Please Don’t Write With A.I.” Writing is not merely the mechanical act of recording thoughts we have already formed. Writing is how we discover what we think.

The struggle matters. Finding the right words exposes holes in an argument. A paragraph that refuses to come together may be telling us that the idea behind it does not hold together. Following a thought onto the page can take us somewhere we did not expect to go.

Artificial intelligence makes it possible to avoid much of that struggle. Give a model a topic, a desired length and a few instructions, and it will produce polished, plausible prose in seconds. A person can publish the result without having developed a genuine point of view, wrestled with the subject or even understood the argument carrying his or her name.

That is a real danger. But it is not the same as writing with AI.

Stephens draws the boundary in the wrong place. He treats using AI and surrendering authorship to AI as though they were the same act. They are not. The meaningful distinction is not between human-written and AI-written prose. It is between intellectual engagement and intellectual outsourcing.

There is an enormous difference between telling an AI model, “Write an article explaining why Bret Stephens is wrong,” and having a sustained conversation with it about where Stephens is right, where his reasoning breaks down and why the distinction matters.

In the first case, the user delegates the argument. In the second, the user develops it.

A serious AI-assisted writing process may begin with a human thesis, experience, instinct or disagreement. The model can test that premise, identify counterarguments, point out unsupported claims, suggest a different structure or offer language that helps the writer see the idea more clearly. The human then judges, rejects, redirects and rewrites.

Sometimes the most valuable thing the AI produces is wrong. Its mistake forces the writer to explain why it is wrong, and that explanation clarifies the argument. Sometimes it returns something superficially elegant but hopelessly generic. Recognizing that requires taste and an understanding of one’s own voice. Sometimes it finds a weakness that the writer would rather ignore. Addressing it makes the finished piece stronger.

That is not disengagement. It can be an unusually intense form of engagement. The writing process has become a conversation, but the human remains responsible for where that conversation goes and what eventually appears under the human’s name.

Simply editing an AI-generated draft does not automatically qualify as authorship. If someone requests an article, changes a few words and publishes it, that person has delegated most of the labor and probably much of the thinking. Putting a human finger on the final product does not make the process meaningfully human.

But consider someone who develops the thesis, supplies the original perspective, argues with the model, verifies the evidence, rejects passages that sound false, restructures the narrative and accepts responsibility for every published word. That person has not become disconnected from the work. The person may be more intimately involved with it than someone who typed a first draft alone and gave it only a cursory edit.

We have heard versions of this argument before. Search engines were going to make us stupid because we would no longer remember information. In one sense, they did change memory. We became better at remembering how to find information and less dependent on retaining every individual fact.

But Google did not eliminate thinking. It changed the allocation of cognitive effort. A lazy user could grab the first search result and repeat it without understanding it. A capable user could compare sources, explore unfamiliar territory and synthesize more information than would previously have been practical.

The tool created opportunities for both intellectual laziness and intellectual expansion. The outcome depended on the user.

AI raises the stakes because it does not merely retrieve information. It can return an answer, an argument and a finished-sounding piece of prose. Its fluency can conceal shallow reasoning. The output may sound authoritative before the user has earned any authority over the subject. That makes it easier to confuse generated language with developed thought.

The response should be AI literacy, not AI abstinence.

Stephens uses the example of an AI-assisted wedding toast. Certainly, someone could supply a few biographical facts and ask a model to manufacture three minutes of synthetic sentiment. The result might be polished, moving and emotionally hollow.

Now imagine a father who supplies actual memories, explains why they matter, struggles with what he wants his daughter to know and rejects every phrase that does not sound like him. Perhaps AI helps him organize those memories or express feelings that he has never been particularly good at putting into words.

Has the technology replaced his love, memories or intent? Of course not. It has helped him translate them into language. The authenticity comes from the human source, judgment and emotional responsibility, not from a requirement that every sentence be constructed without assistance.

None of this means AI collaboration is inherently enriching. It requires discipline. The writer must verify factual claims, resist generic prose, recognize when the model is merely echoing the premise and remain willing to discard an entire draft. The writer must understand the subject well enough to know when the machine is wrong and be honest enough to reconsider when it exposes a legitimate weakness.

Most importantly, the named author must accept full responsibility for the result. “The AI wrote it” cannot become an excuse for an error, a distortion or an argument the author cannot defend.

Calculators changed the arithmetic we perform in our heads. Search engines changed what we remember. Navigation systems changed how we find our way. Each tool weakened certain habits while allowing people to operate at a different level. AI will do the same to writing, but on a much greater scale.

The challenge is learning which parts of the process can be shared with a machine and which must remain human. Judgment, intent, lived experience, taste and moral responsibility cannot be delegated. AI can help construct an argument, but it cannot decide what deserves to be said. It can imitate a voice, but it cannot possess the life that gave that voice meaning.

So I would offer a different rule: Never publish anything created with AI that you have not thought through, made your own and accepted responsibility for.

I am not begging people to stop writing with AI. The technology is here, and refusing to use it will not preserve human thought. I am begging people to learn how to use it without surrendering the thinking that makes writing worth doing.