It brings all the economists to the yard (and they’re fighting)
| Bruno Moreira |
For the past few months, I’ve been drowning in macroeconomic problem sets so abstract they almost felt theological. I’m writing this text to push back against that experience, and in the process to think about our current AI revolution. Everyone is staring at the future like it’s a diamond held up to the sun: dazzling, promising, impossible to look away from. But stare long enough and you could go blind.
So, I decided to reach up, grab it, so that with you, dear reader, we take a magnifying glass and look at the diamond up close. Let’s touch it, feel its edges, and bring our eyes uncomfortably close to its inside.
The AI jobs debate has collapsed into a shouting match. Optimists chant, “We’ve always made new jobs.” Pessimists declare “that’s it! The machines are coming for us. We will all starve and die.” Both sides stand behind their historical charts while they fire. Some shots hit; others miss by far.
Somewhere deep into long hours of studying, grinding through problem sets on dynamic models, technology shocks and price rigidity, around 2 or 3 am, my coffee supply already ran out so I couldn’t get a new kick, as slumber befell my tired eyes, I suddenly realized that economists had already built the exact framework we need. It’s a 25-year-old academic dispute between two schools of thought: Real Business Cycle (RBC) models and New Keynesian (NK) models. One predicts that better technology increases employment. The other predicts the opposite. They’re both right. They’re just talking about different time horizons, and that distinction changes everything about how we should respond to AI.
The Magnifying Glass: A Brief Detour through Macro
(Feel free to skip ahead if you already caught up with your Galí and Kydland.)
The RBC camp, associated with economists like Kydland and Prescott, builds its models on the assumption that markets work, or at least they clear quickly. Prices are flexible, wages adjust quickly, and when technology improves, the economy efficiently reallocates. In this world, a positive technology shock, call it ‘A’, is undoubtedly good. Workers become more productive: each hour of labor (N) produces more output (Y). Firms compete for workers, wages rise, labor supply increases, and the economy expands. Everyone wins.
Written very crudely in macro shorthand:
The New Keynesian (NK) camp agrees with the long-run logic but ruins the party by insisting on a crucial friction: prices and wages don’t adjust instantly. They’re „sticky,“ locked in by annual salary contracts, multi-year agreements, long-term supplier deals, and the simple reluctance of firms to reprice their menus every week, and customers accepting that. In a world with sticky prices, a technology improvement produces an uncomfortable result. In a landmark 1999 paper, Jordi Galí showed that when you isolate genuine technology shocks in the data, employment actually falls in the short run. In the same shorthand, the NK story:
Okay, if what you read until here was a bit indigestible, let’s talk about milkshakes.
Imagine walking into your favorite milkshake shop with your buddies and ordering your favorite strawberry-banana shake. You take a sip and immediately notice it: the shake tastes better. It’s thicker, sweeter, richer. Something clearly improved.
If your buddies happen to be Kydland and Prescott (RBC team), they’re thrilled. “This is great,” they say. “More output. Higher quality. Bigger shake.” End of story. But if you invited your other nerd friend, Galí (NK team), he wouldn’t be cheerful at all. He’d grab the milkshake from your hands, call the barista over and asks an annoying question: “What changed in the recipe?”
Strawberries, in this story, are labor: expensive, fragile, and easy to miss once they dissolve. Bananas are cheaper inputs, but chunkier, visible, easier to count. Galí dumps the milkshake onto the counter and starts sorting through the goo and counting banana chunks. There are more bananas. Fewer strawberries. Sure, the shake tastes better. It’s bigger and cheaper to make. But it uses less labor. That’s the New Keynesian; in the short run, better technology can raise output and destroy jobs.
Are you with me? Now that we’ve got our magnifying glass, let’s go back to the diamond. And yes! Keep up with the analogies. When you’re studying macroeconomics, you’ll need every one of them.
The Front Side Of The Diamond: How Good Technologies Kill Jobs (For Now)
Imagine Alice, Bob, and Carol. They’re all software engineers at the same company, each earning $120,000 a year. Their salaries are locked in until the annual review in December. In October, the company adopts an AI coding assistant. Overnight, Alice becomes three times as productive. She can now do in a week what used to take three weeks.
In the RBC world, the story is simple. The company immediately expands its product three-fold to exploit its new capacity. Bob and Carol are busier than ever building new features. Wages rise. Everyone prospers.
In the real, “sticky-price” world, the company looks at its order book. It has the same contracts, the same clients, the same amount of demand it had in September. It doesn’t need three times the output. It needs the same output at one-third the cost. Bob and Carol are laid off. Their combined $240,000 salary becomes profit.
Then, the cascade begins. Bob and Carol cut their own spending since they are now unemployed. The restaurants they use to go lose them as customers. The landlord tightens up on a lease renewal. Local tax revenue dips. And here’s the kicker: the company’s own clients, watching layoffs spread through the industry, get nervous and reduce their software budgets. Now Alice’s job is under pressure too.
This slows the economy because income falls before demand has time to catch up. Until a new equilibrium is reached (one where new jobs, new markets, and new spending absorb the productivity gain), the economy runs below its potential. This isn’t a fable or just a theory. It’s textbook New Keynesian economics, and it’s playing out right now in the tech sector. Many AI‑layoff firms are profitable. They’re producing the same output with fewer workers while aggregate demand lags behind.
And if you look closely, the problem runs deeper than price stickiness. Prices may adjust in three to five quarters, as in the Calvo model (another friend who just joined our guys for some Milkshake). Human capital doesn’t. Skills take years to retrain. Degrees take longer. Institutions evolve over decades. The friction Galí identified at a quarterly scale is operating, with AI, at a generational one.
This is the front face of the diamond. The part that catches the light first and cuts deepest on contact.
The Backside Of The Diamond: The Optimists Aren’t Wrong Either
Before you sink into despair, here’s where the RBC economists make their crucial point and they’re right. Think about what happened to farmers. In 1800, roughly 90% of the West worked in agriculture. By 1900, it was 40%. By 2000, it was under 2%. Did we have 88% unemployment? Obviously not. We built factories, then offices, then hospitals, film studios, software companies, and data centers. Each wave of agricultural productivity didn’t eliminate work but freed human effort to do things that only became possible because we weren’t all growing food.
This is what economists call the “lump of labor fallacy”: the mistaken belief that there’s a fixed amount of work to be divided up like a pie. In reality, as technology makes some tasks cheaper, it expands what’s possible and creates entirely new forms of demand. A farmer in 1880 couldn’t imagine a job called “UX designer” or “supply chain analyst,” but those jobs exist precisely because productivity made society richer.
This is where I want to leave a small manifesto about something that has always annoyed me in economics class. Economics loves to start with “resources are scarce relative to human wants” like it’s some eternal law, but it’s really a snapshot pretending to be physics. What counts as “scarce” is not stable: technology, organization, and shocks keep changing the menu of what even qualifies as a resource, and our wants evolve right along with it.
The real problem isn’t how to allocate fixed scarcity; it’s that economics often freezes the world to make the math behave. Scarcity is a moving target we keep redesigning, sometimes eliminating it if convenient, sometimes replacing it with a new bottleneck.
Although I will be the first to admit that if it weren’t for log-linearization techniques I would really be damned. But let us call it for what it is: We just cannot know everything. Our most beautiful math models will inevitably misrepresent reality, because we are always taking aspects away from reality to explain it. Treating scarcity as the starting axiom isn’t neutral. It sidelines the larger forces of adaptation and invention. That matters for how we think about AI.
Okay, now that I took that out of my chest, let us wrap up lustering this side of this diamond we are contemplating. The Solow growth model (another buddy joining our Milkshake crew), the powerhouse model of long-run economic analysis, formalizes this intuition. Technology doesn’t just replace labor but expands the economy’s production frontier. As productivity rises, we can afford things we previously couldn’t, which create new industries, and thus, new jobs. In 2070, our grandchildren will have occupations we cannot name, in industries that don’t exist yet, solving problems we haven’t yet identified.
This optimism is probably justified, but how do we get there since right now people are losing jobs and the job market seems to really suck?
The Bottom Side Of The Diamond: Around 40 Years In The Desert
In the Bible, after God finishes dismantling the Egyptian Economy, the Hebrews are promised a new land. If you look at a map it isn’t that far. And yet, they managed to get lost wandering in the desert for 40 years until they got there.
This is where things get uncomfortable for us. The Solow growth model predicts that economies converge to new equilibria at a speed of roughly 2% per year. That sounds reasonable until you do the math. The implied half‑life of an economic transition, that means the time it takes to close half the gap between the old equilibrium and the new one, is around 35 to 40 years.
That is an entire generation living inside the adjustment, paying the cost of someone else’s productivity boom. And that 35-year estimate assumes ideal conditions and its results describes aggregates: functioning labor markets, easy geographic mobility, accessible retraining, rational actors with perfect foresight. Real-world transitions are slower, and the lived experience of individuals is often worse.
As this was not already bad enough, this average hides enormous variation. If generative AI hits in 2026, the software engineer who pivots to prompt engineering may adjust in six months. The lawyer displaced from document review might need years. The truck driver replaced by autonomy may face a decade of precarity. All of these professions are statistically unlikely to recover their previous wages.
The benefits of AI arrive broadly and immediately: cheaper goods, faster services. The costs are concentrated, delayed, and borne by specific people who had no say in the adoption. This is a policy choice about who we ask to wander in the desert so some of us can arrive sooner on our promised land.
The Inner Side Of The Diamond: Why „We’ve Survived Before“ Is The Wrong Argument
The most common optimist argument is also the most intellectually fragile one: „We survived the tractor, the loom, and the assembly line, so we’ll survive AI.“ It sounds reasonable, until you look at it through the lens of the Lucas Critique.
The Nobel laureate Robert Lucas observed that using historical relationships to predict the future breaks down when the underlying conditions change. Past correlations are generated by people behaving under specific institutions, technologies, and incentives. Change those conditions, and the old relationship might vanish.
That’s exactly the problem here. Previous automation happened over decades. The tractor took generations to fully displace agricultural labor. Most past technologies were narrowly adopted, replacing specific physical tasks while leaving cognitive work largely untouched. AI is fundamentally different: it diffuses faster, affects a much broader set of tasks, and arrives in an economy with a radically different financial and institutional structure than in the 1950s.
The fact that we eventually adjusted to the tractor tells us very little about whether we’ll adjust smoothly to AI. The conditions that produced the old outcome no longer hold. The correlation breaks. This is the inside face of the diamond, the part you don’t see until it’s too late.
The Top Side Of The Diamond: What Economics Actually Recommends
The policy conclusion that follows from this framework is more nuanced than either the techno-optimists or the doomers believe it to be.
From the RBC side: “Don’t slow the technology!” Trying to put brakes on AI adoption to preserve existing jobs is the economic equivalent of banning the tractor to protect farm labor. It would work, briefly, and at enormous cost to future prosperity. The long-run gains from AI are real and large. Ignoring them to protect specific job categories is a terrible trade.
From the NK side: “Aggressively manage the transition!” The market will not solve this on its own. People cannot borrow against their future post-retraining income; credit markets don’t work that way. Laid-off workers with mortgages, children, and depleted savings cannot smooth their consumption over a multi-year retraining and adjustment periods. This is a textbook market failure, and it calls for a textbook policy response.
In practice, that means extended unemployment insurance that doesn’t expire after six months. Publicly funded retraining programs that are well-resourced. Wage subsidies for workers entering new industries at lower initial pay. Healthcare decoupled from employment so that job transitions don’t become medical catastrophes. And serious investment in the other aspects of the economy that make us human and give us more dignity (not just machines and algorithms, but the human capital, public infrastructure, and institutional quality), which determines how quickly the individual person, and ultimately entire economies, can absorb disruption.
The goal is to shrink that 35-year half-life and to make the adjustment fast and fair. Because if I’m remembering the Bible correctly, a lot of people died in that desert, including Moses, the leader who started the journey, before anyone reached the promised land.
The Whole Diamond: Let it shine bright, but…
The pessimists are right about the next ten to twenty years. The optimists are right about 2070. The problem is surviving the middle.
If you are reading this, there is a great likelihood you are in the AI-augmented camp. Your skills complement the technology rather than compete with it. The transition looks manageable from where you sit. But your Uber driver, your lawyer, people who create the tv-series you watch, and the radiologist reading your X-rays may have a categorically different perception. That asymmetry, between those who gain immediately and those who pay the transition cost, is the entire political and moral problem of the AI moment.
Macroeconomic models don’t tell us whether to be afraid of AI. Anyone claiming otherwise is overstating their case. What the models do tell us is where to look. Technology itself isn’t the variable that matters most. The core variable is speed of adjustment: how quickly wages, skills, institutions, and social safety nets can catch up to the algorithm.
History says that gap closes… eventually. History also tells us that the people living through the closing don’t get to skip it.
Let us be kind to ourselves and now that we’ve examined the diamond from every side, the least we can do is handle it with care.
Author contact information:
https://www.linkedin.com/in/moreirabruno1
References
Barro and Sala-i-Martin (2003): Economic Growth, Second Edition, MIT Press.
Galí (1999): „Technology, Employment, and the Business Cycle: Do Technology Shocks Explain Aggregate Fluctuations?“ American Economic Review 89 (1): 249–271.
Galí (2003): “New Perspectives on Monetary Policy, Inflation, and the Business Cycle,” Advances in Economic Theory, edited by: M. Dewatripont, L. Hansen and S. Turnovsky, Vol. 3, 151-197, Cambridge University Press.
Galí and Rabanal (2004): “Technology Shocks and Aggregate Fluctuations: How Well Does the RBC Model Fit Postwar U.S. Data?,” NBER Macroeconomics Annual 19, 225-288.
Galí (2015): Monetary Policy, Inflation and the Business Cycle, Second Edition, Princeton University Press.
Galí (2018): “The State of New Keynesian Economics: A Partial Assessment,” Journal of Economic Perspectives 32(3), 87-112.
Kommentar verfassen