A friend told me last week he's terrified AI will take his job. I asked him one question: "What do you actually mean by having a job?"

He paused. And honestly, that pause is where everything starts.

Two Fears Wearing the Same Mask

Let's be straight about this. "Having a job" carries two completely different meanings, and we keep pretending they're one thing.

If having a job means being useful to society — teaching someone, building something, caring for a neighbor — then AI is no threat at all. Machines handling invoices doesn't stop you from doing meaningful work. If anything, it clears the way.

But if having a job means getting a paycheck so you don't starve — then the problem you're describing existed long before any language model. The problem is the distribution system. It was always the distribution system.

And the reality is, those two fears belong to two very different classes of people.

A warehouse worker fears losing his job because his children won't eat. An investor fears workers losing their jobs because nobody will buy his product. These are not the same fear. The first is existential. The second is structural. And the distance between them — that is the class conflict. As Claus Offe and Helmut Wiesenthal argued in 1980, capital and labor are not symmetrical opponents — the employer can wait, diversify, and replace; the worker cannot split their livelihood across multiple lives.

The Paradox Nobody Talks About

In 1943, Michał Kalecki laid out something so uncomfortable that mainstream economics still pretends it doesn't exist. In a ten-page essay in the Political Quarterly, he showed that capitalists oppose full employment even when it would increase their own profits. Why? Because — and I'm paraphrasing only slightly — discipline matters more than money. If workers can't be fired, they can't be controlled. The threat of unemployment isn't a bug in the system. It's a feature.

Read that again. The owning class actively works against conditions that would make them richer, because losing control over workers is a worse outcome than losing profit. Kalecki's exact formulation: business leaders appreciate "discipline in the factories" and "political stability" more than profits, because "their class instinct tells them that lasting full employment is unsound from their point of view." (Full text here.) Now replace "full employment" with "AI abundance" and the logic holds perfectly.

Nobel laureate Daron Acemoglu and Simon Johnson arrive at a strikingly similar conclusion eighty years later in Power and Progress: the question is not whether AI can augment human work — it can — but that the tech industry's business model systematically pushes toward automation instead, because replacing workers is more profitable for capital than empowering them. Acemoglu's 2024 NBER paper puts it plainly: AI is predicted to widen the gap between capital and labor income. The direction of technology is a political decision, not a natural law.

The Efficiency Trap

Here's where it gets absurd. Keynes predicted in 1930 that technological progress would deliver a 15-hour work week within a century. David Graeber looked around in 2018 and asked: so why are we working more? His answer — bullshit jobs. Administrative layering, corporate compliance theater, middle management supervising people who don't need supervision. Productivity gains didn't free us. They generated meaningless work to keep the machine of employment running.

Think of it as a Jevons paradox of labor: every efficiency gain, instead of reducing the total work required, creates new demands that absorb the surplus. We got more productive and more busy at the same time. AI is about to push that contradiction to its breaking point.

Aaron Benanav offers a crucial corrective here. In Automation and the Future of Work, he argues that the declining demand for labor is not primarily caused by automation at all — it's a symptom of capitalist stagnation, of global overcapacity and decelerating productivity growth. (The original argument appeared as a two-part essay in New Left Review.) The "robots are coming" narrative, whether told by Silicon Valley optimists or leftist utopians, obscures the real structural crisis underneath. If Benanav is right, AI doesn't cause the problem — it accelerates a dynamic that was already in motion.

Who Consumes in a Jobless Economy?

But there's a deeper crack in the foundation. If people don't earn, they can't spend. If they can't spend, markets collapse. Agnes Heller saw through this decades ago. Drawing on Marx's early manuscripts, she argued in The Theory of Need in Marx that capitalism reduces all human needs to the need to possess — and then fails to satisfy even that for most people. The real needs — community, meaning, creative work, personal growth — are qualitative. The market can't measure them, and money can't buy them.

So what are the proposed fixes? Universal Basic Income gives everyone cash. Sam Altman's Universal Basic Compute gives everyone a slice of AI capacity. Both sound progressive. But honestly speaking, both leave the underlying architecture untouched. You still need to buy your life on a market controlled by someone else. Karl Polanyi's history of the Speenhamland system in The Great Transformation — England's 1795–1834 experiment in supplementing the poor's income from public funds, effectively a proto-UBI — is instructive: employers simply lowered wages knowing the subsidy existed, and the system institutionalized poverty rather than eliminating it.

Piero Sraffa demonstrated in his 1960 Production of Commodities by Means of Commodities that distribution isn't determined by economic law — it's determined by political power. Given any set of technical conditions, the split between wages and profits has one degree of freedom: it is a trade-off settled by struggle, not by "marginal products." And John Roemer goes further that exploitation doesn't happen at the factory floor. It originates in the unequal distribution of productive assets. If you applied his test to AI — asking whether ordinary people would be better off with their fair share of AI infrastructure — the answer is obvious. The current arrangement is extraction, plain and simple.

Back to the Pause

My friend is still pausing. But now I think the answer is clearer.

You don't want a job. You want a good life. You want to do something that matters, surrounded by people you care about, with enough security to sleep at night. None of that requires a paycheck from a corporation. All of it requires that we stop confusing employment with purpose.

Marx wrote in his Critique of the Gotha Programme that "any distribution whatever of the means of consumption is only a consequence of the distribution of the conditions of production themselves" — you cannot change who gets what without changing who owns what. It's time to rediscover forms of association that aren't companies. Cooperatives, commons, mutual aid networks, community workshops — structures built around shared needs, not extracted profit. The means of production are changing. The question is whether we'll change who owns them, or just keep debating how to redistribute the crumbs.

AI isn't the problem. It never was. The problem is an economic system that can only distribute dignity through a paycheck — and is now running out of paychecks to hand out.