Every economic model has a room it never visits. In this one, it’s the bread line.
The equations are tidy. Compute grows, machines take over the work, output climbs and wages flatten out at whatever it would cost a computer to do the same job. Somewhere outside the model, millions of people are standing in a line with no paycheck, and the model has nothing to say about them. I spent 33 years in technology, and I’ve watched plenty of plans that worked on paper meet people who didn’t read the paper.
I read a paper this week that is all equations and no bread line. Read it, because its author is a serious economist and because it says something blunt about writers. Then argue with it.
What does Pascual Restrepo’s “We Won’t Be Missed” paper say?
Pascual Restrepo is an economist at Yale. His paper, “We Won’t Be Missed: Work and Growth in the AGI World,” is a chapter for a National Bureau of Economic Research volume on the economics of AI, and NBER released it as a working paper. Fortune covered it in April under a headline about jobs that aren’t worth the trouble to automate.
His starting point is AGI, artificial general intelligence: a world where computers can do all economically valuable work, given enough computing power. He splits that work into two kinds. Bottleneck work is the work an economy can’t grow without: food, shelter, energy, infrastructure, science, logistics, defense. Accessory work is everything the economy can keep growing without.
His conclusion is stark. All bottleneck work eventually gets automated. People can still hold jobs, but their wages stop at the cost of the computing power it would take to replace them. That cost keeps falling. Labor’s share of the economy heads toward zero. “The economy keeps growing; we stay in place,” he writes. He ends with the line that gives the paper its title: “If tomorrow half the population stopped working, no one would notice. In the AGI economy, we won’t be missed.”
I don’t buy it.
How Restrepo sorts work in an AGI economy, using the examples from his paper.
Is writing “accessory work” in an AGI economy?
In his model, yes. His list of possible accessory work includes arts and crafts, performing for others, hospitality, fine dining, design, customer support, judicial work, religious and civic groups, and literature. He even puts academic economists on it. I appreciated that.
For writers, that cuts both ways. On the plus side, AI has little reason to take over accessory work, because it doesn’t drive growth and isn’t worth the computing power. Somebody will still write books. The pay is the problem. In his model, accessory work survives at a wage that stays capped, never grows with the economy, and can trade at a discount.
You can only replace so many people before they get angry enough to stop you.
He ends with questions about meaning. Would people keep working when their work makes no economic difference? Would work still feel meaningful? He also floats two ways to share the income that computing power generates: universal dividends, or treating compute as a public resource the way some countries treat land or oil.
The bread line the model leaves out
I think the paper was written from an ivory tower. Restrepo is looking at AI. He isn’t looking at real people living real lives.
Look back through history at what happens when large numbers of people can’t afford food or water. They revolt. In 1789, a bad harvest had pushed the price of bread in Paris beyond what working families could pay, and France got a revolution. In 2010 and 2011, world food prices hit record highs, and protests spread across North Africa and the Middle East. Hunger is the root cause of a lot of revolutions, maybe most of them.
Machines have set it off before. In 1811, English textile workers watched new machinery cut their wages in a year when bread was expensive. They called themselves Luddites, and they broke into mills at night and smashed the frames. Parliament made machine-breaking a capital crime, and the government sent thousands of soldiers to the factory towns.
The word Luddite became an insult. The anger behind it was about feeding a family.
So run Restrepo’s line forward. If half the population stopped working, somebody would notice. Half the population would notice, and they’d be furious. They’re getting angry now, and we’re nowhere near half. In a Pew Research survey in February 2025, 52 percent of American workers said they were worried about how AI will be used in the workplace. Only 36 percent felt hopeful.
Restrepo sees part of this. He writes that the transition will feel uncertain and even unfair to workers. Some people will watch their wages drop to whatever the computer costs. Others, “by a sheer act of luck,” will earn a big premium because their job happened to be among the last anyone figured out how to automate.
Picture that from the street. Your neighbor keeps a good salary because his job was hard to automate. Yours was easy, so your paycheck shrank to the price of a server rack. Neither of you did anything to earn the difference. How long do you think people will accept that?
Not long.
You can only replace so many people with AI before they get angry enough to stop you. Hunger has toppled more governments than any machine. – Richard LoweShare on X
I see that anger as the gating factor on how far AI spreads. You can only replace so many people before you make so many of them so angry that they won’t tolerate it anymore. Any model of the AI economy that leaves out the people who lose their paychecks is missing the variable most likely to break it.
Why are writers so angry about AI?
Writers see AI replacing them, and on one level they’re right. The traditional way of writing, one person alone with a blank page doing every step by hand, is being replaced by AI and will keep being replaced.
Anger won’t change that. You can hold your breath until you turn blue and it won’t change. Writers who attack other writers in Facebook groups won’t change it either. I’ve written a whole series about why writers are angry about AI and how ugly the pile-ons get, and none of that anger has slowed the technology down by a day.
Writers who keep working the old way, refusing to touch the tools, are the ones Restrepo’s model describes. Their work gets cheaper to replace every year, and their pay follows it down.
Can writers who use AI as an assistant keep working?
Yes. I use AI as an assistant. It organizes my research, keeps my follow-ups on track, helps me outline and lets me test ideas at two in the morning. I still do the writing and make the decisions about what a book says and how it sounds.
If you work that way, you grow. You write more, finish more and keep your own voice on the page. You’ll go on being a writer, and you may do better than you ever did alone. I made the longer case in augmented beats replaced, and the same logic sits behind the three categories in whether AI will replace your work.
Restrepo’s model has no slot for a writer who is part human judgment and part machine speed. That writer isn’t competing with the compute. That writer owns it.
Here’s how I use it. AI helps me plot, bounce ideas around, outline, build characters and check pacing. It drafts the grunt work when I want it to. I direct it the way I’d direct a fast, tireless research assistant, and I do the creative work myself. Whoever types the final words, the book gets faster and better.
I think AI is setting off a burst of creativity by knocking down barriers. One person, or a small team, can now build what used to take a building full of programmers. Books work the same way. An author with a story and no time can finish it. An author with ADHD can keep the business side running. Writers fighting AI tooth and nail don’t understand the field they’re standing on, and an economist modeling it from Yale doesn’t understand the people standing on it.
I’m not worried about the writer who adapts.
You can set up the same thing for yourself in a week. Start with the work you hate.
What happens to writers who let AI write everything?
They run into the other wall. If you write everything with AI and put out slop, what’s the point? I don’t think it will do very well, because people are catching on to slop, and they don’t like it.
I don’t like it either.
I’m about ready to quit YouTube. By my count, something like 70 percent of what it shows me now is AI slop, made by somebody who typed a couple of prompts and called the result a movie. I don’t mind a good AI-made video. I mind the junk. My LinkedIn feed is worse, mostly AI-generated ads for worthless products that turn out to be cons.
I don’t care what tool a writer uses. I care whether the book is good. Does the horror horrify? Does the fantasy create wonder? Do readers feel like they’re inside the story? A book generated from a one-line prompt has no soul, because the soul lives in the little human details an author puts in on purpose. Nobody puts those in with two prompts and a publish button.
Readers behave the same way viewers do. They’ll put up with a lot, until they won’t, and then they leave. A writer who floods the market with machine-made books is betting that readers can’t tell the difference. More of them can every month. Somebody is counting the AI slop, and the readers are counting too.
My take
Restrepo’s math may be right, and his people are missing. Hungry people revolt, and they’ll put a ceiling on how fast AI replaces anyone. Writers have three choices. Refuse the tools and watch the old way of working disappear. Hand AI everything and join the slop readers already resent. Or use it as an assistant, keep your judgment and your voice, and write more and better than before. Only the third one has a future.
Somebody will notice
If you’re a writer worried about where this goes, pick the third path this month. Try AI on the work you hate, the research, the outline, the follow-up email, and keep the writing for yourself. If you want help setting that up for your own book, my writing coaching covers it, and the AI and writing hub has the rest of what I’ve written on the subject.
Restrepo says we won’t be missed. The people standing in the bread line will disagree, loudly, and the economists will hear about it.
If you hand AI everything, you lose the judgment that made you worth hiring. I wrote this about how that loss happens one reasonable decision at a time.
If the job is already gone, this is my honest guide to the search, starting with the first two weeks nobody prepares you for.
