AI didn't make that decision. A person did.

A friend sent me a paper last week, and I really enjoyed reading it.

It's by Maria Elena Mejia, an economist working on what she calls human-centered economics, and it takes aim at a sentence most of us have stopped questioning: AI is taking the jobs. We read it in the headlines, we repeat it at dinner, and we've started to treat it the way we treat the weather — as something that happens to us rather than something we decide.

Her argument is that this is precisely backwards. And to make the point, she reaches back more than two centuries.

The mistake we already made once

In 1798, Thomas Malthus looked at population and food supply and concluded that famine was inevitable. The maths seemed airtight: people multiply faster than crops. What he could not see — what he had no way of seeing from where he stood — was that technology would expand the entire production frontier. Fertiliser, mechanisation, new crops. The fixed pie he was reasoning about wasn't fixed at all. He mistook the limits of his own moment for a law of nature.

Mejia's claim is that we are doing the identical thing with AI. We look at today's firms, today's quarterly incentives, today's habit of measuring success by headcount removed, and we project forward to mass displacement as destiny. But the thing we're treating as an iron law — technology arrives, people lose their jobs — is not a law. It's an arrangement. We built it, and we can build a different one.

She points out that even Keynes saw further than we tend to. He predicted the productivity gains would arrive (he imagined 15-hour work weeks within a century, and the productivity did arrive). What he got wrong was assuming the gains would be shared. Instead, "technological unemployment" quietly became a measure of success — cutting labour got baked into our definition of progress itself.

The machine isn't the one deciding

Here's the part of her argument I keep returning to. We talk about AI as if it has agency — as if it surveys the org chart and decides who stays. Mejia walks through how the technology is actually built — transformer models, pattern recognition trained entirely on past data — to make a simple but telling point: an AI system has no access to the future, forms no goals of its own, and decides nothing about anyone's livelihood. It is, structurally, a machine for finding patterns in what has already happened.

So when a layoff is announced "because of AI," that phrasing is doing real work. It's moving responsibility off a person and onto a tool. Every one of those decisions was made by a manager, an executive, a board — humans with names, operating inside institutions that reward some choices and not others. "AI took the jobs" isn't just inaccurate. It's an alibi.

The productivity illusion

If cutting staff isn't technological destiny, why does almost every firm reach for it? Because it looks like efficiency. But Mejia draws a sharp line between cutting costs and genuinely growing productivity. Real productivity means more output per unit of input, or freed-up resources flowing into higher-value work. Sacking people and booking the saving is neither — it's a private gain that quietly exports its costs to everyone else: shrinking consumer demand, eroding tax revenue (the money that funds health, pensions, schools), and mounting pressure on public budgets exactly when displaced people need them most.

She calls this the productivity illusion: firm-level savings dressed up as economic growth, while the wider economy absorbs the bill. And because workers are also customers, a system that systematically reduces its own workforce eventually contracts the demand that funds those same firms. It's a strategy that, taken far enough, eats itself.

Drawing on the institutional economist Douglass North — "institutions are the rules of the game" — she argues the deck is simply stacked toward substitution. Replacing people is fast and legible. The harder path, using the technology to expand what people can do, demands vision, patience and management skill that current incentives don't reward. Germany's Kurzarbeit work-sharing scheme is her counter-example: spread the adjustment instead of concentrating it in layoffs, and you get faster recoveries and preserved know-how. It isn't theoretical. It's documented.

We've built guardrails before

The most hopeful stretch of the paper is also the most concrete. Mejia points out that societies have repeatedly stepped in when a transformative technology started imposing its costs on people who never agreed to bear them — and she lists the precedents:

• The FDA, born when unsafe food and drugs were harming the public, on the principle that you prove safety before deployment rather than compensate for harm afterwards.
• The EU AI Act (2024), the first comprehensive legal framework for governing AI.
• A 2026 court ruling in Hangzhou, China, that an AI-driven dismissal was unlawful — establishing that adopting AI is a business choice, not an act of God.
• California's Executive Order N-6-26 (May 2026), directing the state to assess AI's workforce impact and even explore worker-ownership models.

Different legal traditions, different continents, the same underlying recognition: the costs of a technological transformation can't simply be handed to workers without oversight. Protecting people and sustaining innovation, she argues, are not opposing goals. We keep being told to pick one. The history says we don't have to.

Where I come down on it

I want to be careful not to read the paper as saying nothing will change. It won't stay the same. The economics of this are powerful, and they will force real change through whether we like it or not. Some jobs really will disappear. The honest response isn't to ignore that or to dig in and fight it — it's to work with the change and shape it well.

But here's the thing I keep coming back to. AI raises productivity. Our collective capacity to make things and provide services is going to grow, and grow a lot. So whatever the problem is at root, it can't be that there won't be enough to go round. We will have more, not less.

Which means the real question isn't scarcity. It's distribution — how we share the gains fairly, and at the very least make sure that no one is left behind completely, without food, shelter or healthcare. That's a choice too, and a harder one than it sounds, because it cuts against incentives that reward concentrating the gains rather than spreading them.

This is exactly why work like Maria's matters. We need people looking hard at the mechanics of how we manage this — what the rules should be, what to measure, where the guardrails go. I'm genuinely optimistic here: we could get this right. It isn't guaranteed, and it won't happen on its own, but it's within reach if enough people do their collective best to steer it. That feels worth the effort.

Why this stayed with me

I build a platform for self-knowledge, not an economics practice, so I'll leave the policy detail to the economists. But there's a thread in this paper that runs straight through everything we care about.

When we say "AI took the jobs," we don't only get the economics wrong. We also tell millions of people that what's happening to their working lives is weather — impersonal, inevitable, no one's call. That framing does something to a person. It's hard to adapt to a change you've been told is no one's decision and beyond anyone's reach.

Mejia's reframe gives the agency back. If displacement is a choice, then it can be a different choice — and the people living through it are not merely standing in the path of a storm. That matters for the economy. It also matters for what it does to us to be told, over and over, that the future is something happening to us rather than something we still get a say in.

It's a genuinely fascinating paper, and it deserves a wider read.

---

Maria Elena Mejia's working paper, "The New Malthusian Mistake: Why Are Societies Choosing a Shrinking Economy Over True AI Growth?" (1 June 2026), is available on SSRN — Maria introduces it here. This is my summary of her argument, shared with her permission; any clumsiness in the compression is mine, not hers.