On July 28, 2026, Mark Zuckerberg published a piece in the Wall Street Journal arguing that superintelligence should belong to everyone. His position is that concentrating that much capability inside a few institutions would be worse for safety than spreading it. The values that carried us this far, liberty and open inquiry and free enterprise, are the ones to build on.
I ghostwrite books for a living, so when the CEO of Meta publishes a philosophy of who should be able to make things, I read it as someone whose whole trade is making one kind of thing for other people. I agree with more of it than I expected to. I also think it gets books badly wrong, and the people it’ll hurt are the ones with a real story who come away believing a machine has made that story worthless.
Most of the argument is reasonable. The weak spot is a thought experiment about lawyers. It works exactly as he says in the setting he chose, and it stops working the moment you move it.
What is Mark Zuckerberg’s argument for personal superintelligence?
His claim comes in three parts. Putting capability in individual hands is where prosperity comes from. Invention, and not automation, is what superintelligence is for. And balance of power, spread widely, is a better foundation for safety than a few well-intentioned institutions holding everything.
He makes a point about jobs that most coverage skipped. There’s a balance, he writes, between using AI to automate and using it to help people expand what they can do. If it tips toward automation, the effect on work may be negative. Distribute the capability and he expects more businesses, most of them small.
Then comes the thought experiment. Imagine one person has a superintelligent lawyer. He wins in court whether or not he should, and society is worse for it. Now imagine everyone has one. Justice gets carried out more fairly and more efficiently than it does today.
Where the lawyer argument holds
It works because a courtroom has three features that most situations lack. The dispute is adversarial and bounded. Two sides, one question, one outcome. There’s a decision maker who’s obliged to reach a verdict, and can’t decline on the grounds that both arguments were tedious. And the merits exist independently of how well they’re argued. Somebody either signed the contract or didn’t.
Put those together and Zuckerberg’s conclusion follows. If skill in argument is unevenly distributed, the better argued case wins regardless of the facts. Distribute the skill and the advantage cancels out, leaving the facts to decide. He’s describing the removal of an asymmetry, and in a system with a compulsory judge, removing that asymmetry does move outcomes toward the merits.
I’ll give him that one without a fight. Anybody who’s watched a well-funded litigant bury a smaller one in motions knows the courtroom tilts toward whoever can pay for the better argument, and leveling that would be a good thing.
The same argument on a bookshelf
Now run it again with books.
One person has a superintelligent ghostwriter. Their book is competent, well structured, correctly paced, free of the errors that mark an amateur manuscript. Against a field of ordinary books, it stands out. Zuckerberg’s first case, exactly as described, and I agree with him that this is a bad outcome.
So distribute it. Everyone has one. Every person who wants to write a book can produce a competent one.
What follows isn’t a fairer bookshelf. Competence stops functioning as a signal, because a signal only works when it’s scarce. A reader used to be able to tell a serious book from an unserious one in a page. That no longer works. The unserious book now has the same clean prose and the same tidy structure. The asymmetry has gone, and nothing has replaced it.
That collapse worries me for the people I write for. Somebody with decades of hard-won judgment used to stand out just by getting a well-made book into print. Now that book sits on the same shelf as thousands of polished ones written by people who’ve never run anything, and a reader can’t tell them apart from the first page.
The difference is the judge. A courtroom has one and they must rule. A bookshelf doesn’t. A reader is under no obligation to finish anything or reach a verdict. They can put the book down and never think about it again, and they will, because there are more books than hours.
Why does an attention market behave differently from a courtroom?
Because the constraint sits in a different place.
In court the constraint is capability. One question to settle, somebody obliged to settle it, and the side with the better lawyer holding an unearned edge. Fix capability and you fix the distortion.
On a bookshelf the constraint is attention. The question isn’t which of these two books is better, it’s whether this book is worth an evening when a hundred thousand others were published this year. Doubling the number of competent books doesn’t double the hours anyone has to read them. It divides the same attention across a larger field.
Zuckerberg’s own framing points at this without following it through.
Invention and not automation, he writes, and the number of questions you can ask in a day is limited while the number of things superintelligence can invent is unlimited. That’s right, and it means the bottleneck moves. When production is unlimited and consumption isn’t, whatever governs consumption becomes the scarce thing.
For books, that’s trust. Why should a reader believe this particular person has something worth their evening? Prose quality used to be a rough proxy, and it worked because writing well was hard. It no longer is.
Does AI make writing a book pointless?
No, and I want to be careful here because the doom version of this argument is as wrong as the utopian one.
What has become cheap is the production of competent text. What hasn’t changed at all is the value of having lived something and being able to say what it meant. A model can write clean sentences about a subject it has no relationship with, and any writer who tells you otherwise hasn’t used one seriously. It can’t tell you what it was like to run a company through a period when payroll wasn’t certain, because it wasn’t there.
The people who should worry are the ones whose offer was fluency. If what you were selling was the ability to produce readable prose on a topic, that skill has been commoditized, and the pricing will follow. I don’t have much sympathy for the content mills that built a business on that. They sold words by the pound, and now a machine sells them cheaper.
The people who shouldn’t worry are the ones whose offer was the material itself. A memoir works because of what happened. A business book works because somebody made decisions under real pressure and can explain why. The question of whether a story is worth telling was never a question about writing ability, and distributing writing ability doesn’t touch it.
The fork between automation and assistance
Zuckerberg puts the whole employment question on this fork, and for writing the fork is unusually visible.
Automation is asking a model to produce the chapter. Assistance is asking it to find the three studies you half remember. To check whether a date is right.
To tell you which of two structures a reader would follow more easily. The first replaces the judgment you were being paid for. The second removes the drudgery around it.
I hold that line in my own work and it’s easy to check.
The tools help me find things, organize things, and test things. They don’t produce the prose that goes in the book. That’s a practical position about what a client is paying for. When somebody hires one person’s judgment applied to their story, a chapter written by a model isn’t that, and it doesn’t matter that it reads well.
What gets scarce when everyone can write well
Four things, and none of them are craft.
Direct experience is the first. What it’s like to make a decision with incomplete information and real consequences. No amount of capability generates that, because it’s a history.
No amount of capability generates that, because it isn’t a capability at all, it’s a history.Share on X
Nerve is the second. Most books are worse than the conversation that produced them, because the writer sanded off the part that was interesting. The habit of waiting until you know enough is usually the polite name for not wanting to commit to a position.
Restraint is the third, and unlimited production makes it harder. When generating another chapter costs nothing, the discipline to cut one becomes the scarce skill.
Trust is the fourth. If competence no longer distinguishes anything, readers fall back on whoever they already believe. That advantage accrues over years and can’t be generated on demand, and that’s why it’ll matter more.
What should you do if you’re planning a book in the age of AI?
Write the one only you can write. That advice was always true and always a little glib. It’s now operationally specific, because everything else has been automated and this hasn’t.
In practice that means favoring the material a model has no access to. What you saw, what you decided, what it cost, who was in the room.
A book built on public information about your industry is one a model can approximate. Nothing can approximate a book built on the eleven years you spent inside that industry, and that’s the book I’d tell anybody to write.
It also means the interview record becomes the asset. When I work with a client, every sentence traces back to a conversation on a particular date, and I can tell them which one. That used to be a habit. After a court settled a billion and a half dollars over where text came from, it’s a document with a use.
And it means accepting that the shelf got louder. More books will be published, most will be competent, and almost none will be memorable. That’s a harder market for the average book and no harder at all for a good one, because the thing that made a book good was never in the sentences.
Is widely distributed superintelligence good for writers?
Probably, on balance, and not for the reason the essay gives. Zuckerberg is right that concentration is the greater risk. A world where three companies decide which stories get told is worse than a noisy one, and publishing already ran that experiment with gatekeepers for a century. The noise is a real cost and it’s a smaller cost than the alternative.
But the benefit is that the people who have something to say and never had the craft to shape it now can, and the barrier that kept them out was always arbitrary. Craft was a filter that selected for having learned to write. That correlates weakly with having something worth writing about.
It’s also why I think writing with AI is fine for people who can’t afford a ghostwriter. A tool that gets a real story out of somebody’s head and onto a page is doing good work, and I’d much sooner see that book exist than watch it die because its author couldn’t pay somebody like me. Remove that filter and more of the right books get made, along with a great many of the wrong ones. Net positive, and messy.
Which brings the lawyer back. Zuckerberg is correct that everyone having one improves justice, because a judge has to choose and the merits decide once the skill gap closes.
On a bookshelf nobody has to choose anything. The reader simply leaves, and the only defense against that has ever been having something to say that they can’t get anywhere else.
If you’re weighing up a book of your own, the practical next step hasn’t changed and has become more urgent.
Work out what you know that nobody else does, and build the thing around that instead of around the writing. The AI and Writing Hub collects the pieces on tools, disclosure, and where the line sits, and the ghostwriting service page explains how the interview record works on a real project. The bookshelf is getting louder either way.
Zuckerberg can keep his superintelligent lawyers. I’m worried about the person with a real story who gives up on writing it because somebody told them AI made books pointless. That advice is cruel, because the one thing a machine can’t produce is the thing that person already has, and people are telling them to throw it away.
