Bret Stephens in the NYT: never write anything with A.I.

The interesting thing about a New York Times opinion column that ran this week is not the argument, which is familiar, but the flatness of the refusal. Columnist Bret Stephens tells readers, "Don't use artificial intelligence to help you write. Never let A.I. do your writing for you," and he explicitly folds in school papers, work briefs, speeches at company gatherings and "emails (however perfunctory) to your colleagues or friends." He also tells readers not to let A.I. suggest an opening sentence or a closing paragraph, and not to pretend that editing an A.I. first draft makes the writing yours.

Stephens' real argument is not primarily about cheating, though he calls A.I. writing "unethical when it's a deception." The stronger claim is cognitive. Writing with A.I., he says, is "mentally enfeebling," "an escalator toward a result when you really need to make a daily habit of taking the stairs." Writing, in his framing, is what "compels thought," forcing the writer to subject an idea to "the effort of articulation, the rigor of grammar, the tests of intelligibility and coherence, the inspection of others."

He grounds the alarm partly in a recent exploratory analysis by Dan Sarofian-Butin of 100 EdD dissertations published in 2025 in the field of educational leadership and administration, which he reads as evidence that A.I. is already threading through the scholarly record. Worth taking that specific finding as reported and single-sourced for now, not as settled evidence.

The honest caveat is that this is an opinion column, not an audit. Stephens does not draw a clean line between generative A.I. and the older grammar and spell-check tools most professionals already lean on, and he does not engage with the accessibility case, where A.I. has become a real leveler for students working in a second language or with a learning disability. It also sits awkwardly beside reporting that the New York Times is itself encouraging staff to use A.I. to draft headlines.

The part worth watching is whether the "writing is thinking" frame catches on at the level where policy actually gets written, in schools, law firms, universities and government. If it does, the vendor pitch that A.I. writing tools are a pure productivity win gets harder, and the market for detection and for human-written provenance gets easier.