I. A Lecture in the Lion's Den
Last year, OpenAI's chief executive Sam Altman did something unusual: he invited Dave Eggers to the company to address an audience of roughly two hundred employees.
Eggers was not there to cheer. He is the author of The Circle — the novel that renders Silicon Valley as an open-plan prison. He founded the literary magazine McSweeney's. He built a school to support young writers. He had said publicly that AI-generated text is "pastiche nonsense."
Altman must have had some idea of what he was in for when he invited that man into OpenAI headquarters. But he may still not have expected Eggers, before two hundred employees, to say what he said.
II. "Their Voice Is Stolen from Them"
Eggers's heaviest lines were aimed at writing and education.
The effect of ChatGPT on educators' lives is catastrophic.
OpenAI, he said, has made every teacher's life immeasurably harder. Then he pointed at something more fundamental —
If students are using it to compose... they'll never learn to write. And their voice is stolen from them.
And finally the judgment that has been quoted ever since: a generation that loses the ability to tell its own stories ends up —
silencing an entire generation or two.
Attend to the word: stolen. Not "replaced." Not "assisted." Stolen. Eggers chose a word that carries the smell of expropriation. The voice was yours; it was taken; you may not even have noticed.
III. Writing Is Not the Production of Text. It Is the Organization of Thought.
To hear Eggers's warning rightly, a popular misconception has to go first: that writing is an "output" — labor that turns thoughts into words, which AI can do for you the way a calculator does your arithmetic.
Wrong.
The real value of writing was never in the delivered text. It is in what the process of writing forces you to do. You must press a fog of feeling into a sentence of clear judgment. You must discover that two of your ideas contradict each other. You must admit that what you thought you understood, you cannot actually say.
That process is thinking itself. The text is only thinking's by-product — the footprints it leaves behind.
When AI performs the passage from thought to text for you, what you save was never the typing. What you skip is the stretch of thinking that only slow, solitary writing can force out of you. You believe you outsourced the drudgery. You outsourced the mind.
This is what "the voice is stolen" actually means. Voice is not a matter of style; it is a matter of cognition. A person's voice is a fingerprint of thought, ground out over years of writing, deleting, starting over. Cancel the process, and the fingerprint never forms. A generation grows up fluent — and unable to say anything that is truly its own.
Which also explains why Eggers calls AI-generated text "pastiche nonsense."
The judgment sounds harsh. It is precise. AI's text is never "without style" — on the contrary, it can imitate any style, write more smoothly and more neatly than most people. What it lacks is not form but the living person behind the form. A person's voice is not the sentence patterns they choose; it is the angle on the world grown over decades of reading, of being wrong, of changing one's mind. It cannot be copied by style transfer, because it is not a coat that can be taken off. It soaks through every sentence. It is the person.
AI can produce impeccable text. What it cannot produce is the thing behind the impeccability: a person who has actually thought, and actually changed their mind. So the text reads all-correct and empty. Form complete, voice absent. That is Eggers's "nonsense" — not because something in it is wrong, but because there is no one behind it.
IV. Why "Catastrophic"
Eggers singled out education. ChatGPT, he said, is catastrophic for educators.
To understand the word, shift to the teacher's side of the desk. The core of education was never "delivering knowledge to students." It was forcing students to turn the vague into the clear — an assignment is a mechanism that forces you to organize the fog in your head into words another person can read. What the teacher grades, more than grammar, is the clarity of your thinking.
AI hands over the finished product, and the mechanism snaps. What the student submits is a flawlessly organized text that required him to think nothing through. What the teacher faces is a room of students who appear to know everything and have processed nothing. You can correct the grammar. You cannot correct a thinking that never happened.
That is the weight of "catastrophic." It is not that the workload grew. It is that the core mechanism of education — writing as the forcing-house of thought — has been bypassed. And a bypassed mechanism does not restart itself.
V. A Destruction More Quiet than Job Loss
Eggers calls AI-assisted writing "the biggest tragedy of all." The judgment deserves a pause.
The earlier essays in this series described what AI does to external structures: how it forced the question of ownership onto the table, how it remade pricing and rent, how it disciplines labor. All of these are changes in the relations between people — who owns what, who pays rent to whom, who manages whom.
Eggers speaks of a different quadrant: AI's transformation of a person's relation to themselves.
Job loss, however severe, is an external problem — you lost a job; you can retrain, change tracks; society can argue about redistribution. The atrophy of cognition is another matter. It is not that you lost something. It is that something never grew. A generation that has never truly written will not know the ache of "I thought I understood until I tried to write it" — and that ache is precisely the road thinking must take to mature.
This is a more hidden expropriation than unemployment. Unemployment you can see, rage at, organize against. The disappearance of the voice is silent — you cannot even mourn a capacity you never had. That is where Eggers's fear concentrates: AI does not take away people's voices. It makes the voice unnecessary before it is ever found.
VI. Editorial Note: Both Quadrants Must Be Seen
Placing this reading last in the series was deliberate.
The previous essays took AI apart as a problem of political economy: property, rent, labor. Those analyses are right, and necessary. But stop at that layer and the picture is incomplete — as if the only thing about AI worth caring about is how it redistributes money and power.
Eggers turns the camera ninety degrees. He reminds us that AI is not only redistributing between people; it is changing people themselves. It is reshaping whether a generation can write, can think, has a voice of its own. This change will appear in no report on "the economic impact of AI," because it is not something economic indicators can capture. Yet it may decide, more deeply than any economic impact, what future people are.
The lead essay of this series, "Power in the Architecture," argues that AI's technical architecture embeds class dispositions. Eggers's warning supplies the other side: the architecture also embeds a quieter, gentler discipline — it persuades people that they no longer need the strenuous thinking that only slow, solitary writing can force out. The first is the discipline of power. The second is the discipline of capacity. Two ropes — one around the outside, one around the inside.
The rope of power can be loosened by property correction and design correction (see the lead essay). The rope of capacity has no institutional fix. It asks each generation to choose again whether to keep the inefficient, painful, and for that very reason irreplaceable habit of writing and thinking for oneself.
The significance of Eggers's lecture at OpenAI is not that he scolded AI. It is that he chose the one angle no economic analysis can reach, and said the one sentence no earnings call will ever contain:
Some things, once stolen, you will not even know ever existed.
This piece is based on The Verge's report of July 18, 2026; Eggers's statements are quoted in his original words. Biographical details (McSweeney's, The Circle, "pastiche nonsense") are drawn from the report.