The Ultimate Case for AI Optimism

There is a spectre haunting Silicon Valley, and for once it is not regulation. It is the fear that artificial intelligence, for all its promise, will break the economy. Mass unemployment. Rampant inflation. A new underclass of permanently displaced workers. Dario Amodei, CEO of Anthropic, devoted a substantial section of his recent essay “The Adolescence of Technology” to warning about labour market disruption, predicting that AI could displace half of all entry-level white-collar jobs within five years.

These concerns deserve serious engagement. They are not the product of Luddite ignorance but of genuine uncertainty about a technology advancing faster than any in history. Yet when you trace through the economic logic carefully, the catastrophic scenarios dissolve. What remains is not a story of doom but perhaps the strongest case for technological optimism we have ever had.

The stagflation puzzle

The darkest economic scenario combines the worst of both worlds: high unemployment and high inflation simultaneously. Workers lose their jobs to AI, but somehow prices also spiral upward. This is stagflation, the nightmare of the 1970s, and some fear AI could bring it back.

The problem is that there is no coherent mechanism for AI to cause this. Consider what AI actually does: it expands productive capacity. A language model that can write code, draft documents, or analyse data is not destroying supply like an oil embargo. It is creating more of it. One engineer with AI assistance can now do what previously required three. That is a supply-side expansion, not a contraction.

If AI displaces workers, those workers lose income and spend less. Aggregate demand falls. Meanwhile, supply has expanded because AI is producing more goods and services per unit of input. More supply plus less demand equals falling prices, not rising ones. This is disinflationary, possibly deflationary, but certainly not stagflationary.

Some have argued that certain sectors might stay expensive even as others get cheaper. Housing, healthcare, land. Perhaps AI cannot reduce the cost of living where it matters most. This objection fails on its own terms. If workers are losing jobs en masse, they cannot pay high rents. Landlords cannot extract money that does not exist. Demand for housing falls alongside employment. The same logic applies to any sector: widespread unemployment necessarily reduces demand across the board, putting downward pressure on prices everywhere.

The 1970s stagflation had a coherent cause. Oil shocks reduced supply while demand remained strong, pushing up prices. Monetary policy was too loose for too long. AI is the opposite situation entirely: a massive supply expansion. Mapping the pattern of one onto the other is a category error.

The monetary policy escape valve

Grant for a moment that AI causes significant unemployment. Workers are displaced, demand falls, the economy weakens. What happens next?

Central banks cut interest rates. Borrowing becomes cheaper. Firms invest more, consumers spend more, demand recovers. The unemployed find new jobs in whatever sectors are expanding. This is not a speculative mechanism; it is the basic operation of monetary policy that has worked for decades. The Federal Reserve cut rates aggressively during the 2008 financial crisis and again during the pandemic, preventing what could have been catastrophic unemployment from becoming permanent.

The objection here is that AI displacement might be “structural” rather than “cyclical.” Perhaps the jobs simply do not exist anymore, and no amount of stimulus can recreate them. This brings us to the heart of the matter.

Comparative advantage: the mechanism that protects employment

In 1817, David Ricardo demonstrated something counterintuitive. Even if one country is better than another at producing everything, both countries still benefit from trade, and both countries still have something to produce. What matters is not absolute advantage but comparative advantage: each party specialises in whatever they are relatively best at.

This principle applies with full force to humans and AI. Suppose AI becomes better than humans at every single task. Coding, writing, analysis, design, research. The naive conclusion is that humans have nothing left to offer. The Ricardian conclusion is precisely the opposite.

AI cannot do everything at once. It must allocate its capacity somewhere. Rational allocation means deploying AI where its advantage over humans is greatest. If AI is 100 times better than humans at coding but only twice as good at elderly care, you deploy the AI on coding and leave the elderly care to humans. Not because humans are better at elderly care in absolute terms, but because AI’s comparative advantage lies elsewhere.

This is not a loophole or a temporary reprieve. It is a mathematical necessity. As long as there are tasks to be done and AI capacity is finite, humans will be employed in whatever AI is relatively least dominant at. The market-clearing wage adjusts, and people work.

The only escape from this logic requires assuming that AI capacity becomes essentially infinite and free, so abundant that there is no need to allocate it at all. At that point, scarcity itself is mostly solved, and “unemployment” means something entirely different. We would be in a post-scarcity economy, which is a strange “problem” to have.

The subsistence objection

A sophisticated critic might accept comparative advantage but worry about wages. Perhaps humans remain employed, but the market-clearing wage falls below subsistence. People have jobs but cannot afford to live.

This concern also dissolves under examination. If AI is expanding supply, goods become cheaper. The same nominal wage buys more. A subsistence threshold defined in real terms (food, shelter, clothing) falls alongside nominal wages. The worker earning half as many dollars can still afford the necessities if those necessities cost a third as much.

This is not hypothetical. Over the past two centuries, real wages have risen dramatically even as vast categories of human labour became obsolete. Agricultural workers did not starve when tractors arrived; they moved to factories, and the food produced by tractors became cheaper, raising their real standard of living. Factory workers did not starve when robots arrived; they moved to services, and manufactured goods became cheaper.

The pattern is consistent: technology displaces specific jobs while raising real wages overall. AI is a more dramatic version of this pattern, not an exception to it.

Speed and the transition

Dario Amodei’s concerns, and those of many thoughtful observers, focus less on the equilibrium than on the path. Perhaps the economy eventually adjusts, but the adjustment is so fast and so broad that the transition itself is catastrophic. Social unrest, political instability, a generation of workers left behind before markets clear.

This deserves acknowledgment. Previous technological transitions occurred over decades. Agriculture’s share of employment fell gradually from 90% to 2% over more than a century. AI might compress comparable changes into years. That is genuinely unprecedented.

Yet speed does not break the mechanism. It merely determines how aggressive the policy response must be. If displacement is fast, unemployment rises quickly, and central banks cut rates quickly. The adjustment is bumpy, perhaps very bumpy, but “bumpy” is different from “broken.” The economic logic still holds; it simply operates on a compressed timeline.

Moreover, the breadth of AI’s capabilities cuts both ways. Yes, it can substitute for a wide range of human cognitive tasks. It also means that new jobs created by AI-driven economic expansion can emerge across an equally wide range of sectors. The economy is not searching for one narrow niche where humans retain advantage; it is rebalancing across countless margins simultaneously.

The investment boom

The discussion so far has assumed that monetary policy must actively intervene to maintain employment. There is a more optimistic scenario in which intervention is barely necessary.

If AI raises the marginal efficiency of capital (investments become more productive), then at any given interest rate, more projects are worth undertaking. Firms borrow more, invest more, produce more, and hire people to complement that capital. Demand rises organically, without the central bank needing to cut rates at all.

This is the classic story of technological progress driving growth. The steam engine, electricity, and computing all raised returns on investment and triggered investment booms that created jobs far exceeding those displaced. AI is exceptionally well-positioned to follow this pattern because it augments productivity across nearly every sector simultaneously.

In this scenario, rates might stay flat or even rise (because demand for borrowing is high), the economy grows rapidly, and employment holds steady or increases. Workers earn higher real wages because they are more productive and because the goods they buy are cheaper. This is not a scenario requiring careful management; it is the economy doing what it does when a transformative technology arrives.

What remains to worry about

None of this is to say that AI presents no challenges. There are genuine concerns that survive economic scrutiny.

Distribution is one. Even if the economy grows and employment holds, the gains might accrue disproportionately to those who own AI capital. This is a political question about taxation and redistribution, not an economic question about whether markets function. It is soluble through policy choices we already know how to make.

Transition pain is another. The jobs of many people are going to substantially change. Even if adjustment takes only a few years, those years may be difficult for specific workers in specific sectors. Retraining programmes, income support, and labour market flexibility all matter for how smoothly the transition proceeds. This is a question of how much suffering occurs during the adjustment, not whether adjustment is possible.

Political economy is a third. If the transition generates enough anger and instability, societies might make policy choices that actively prevent adjustment, such as protectionism, excessive regulation, or punitive taxation that kills the golden goose. This is a risk about human decision-making, not about economic mechanisms.

These are real concerns. They are also manageable concerns, familiar from previous technological transitions, addressable through competent policy. They are not civilisational threats.

The ultimate case

Let us now state the optimistic case directly.

AI is a supply-side expansion of extraordinary magnitude. It will produce more goods and services, more cheaply, than humanity has ever managed. Comparative advantage guarantees that humans remain employed in whatever AI is relatively least dominant at, and falling prices ensure that even lower nominal wages translate to higher real living standards. Monetary policy provides a reliable escape valve if demand falls short, and the investment boom may render even that unnecessary.

The catastrophic scenarios, stagflation, mass permanent unemployment, civilisational disruption, lack coherent mechanisms. They rely on assuming away the forces that have allowed human economies to absorb every previous technological shock. Tractors, factories, computers, the internet: each was predicted to end work as we knew it, and each instead raised living standards while creating new categories of employment.

This is not naivety. It is the accumulated evidence of two centuries of technological progress, combined with the basic logic of how markets function. AI is dramatic, fast, and transformative. It is not magic, and it does not repeal economics.

The coming years will bring challenges, disruptions, and difficult adjustments. They will also bring extraordinary gains in productivity, health, scientific discovery, and material abundance. On the other side of this transition lies a wealthier, more capable, more flourishing human civilisation.

That is the ultimate case for AI optimism. Not because the transition will be painless, but because the mechanisms work, the logic holds, and history suggests we are more resilient than we might think.