This past week has been a watershed moment in the AI and employment debate. Citigroup’s strategists, led by Dirk Willer, published a note warning that AI may eventually lead to higher unemployment and deflation, even while acknowledging that the timing remains unclear. Coming on the heels of the viral Citrini Research report, which painted a fictional but disturbingly plausible picture of a 2028 economic crisis triggered by white-collar AI displacement, and Jack Dorsey’s announcement of a 40% workforce reduction at Block, it feels as though the conversation has shifted. We are no longer debating whether AI will displace workers. We are debating how fast, how broadly and what on earth we plan to do about it.
The optimists have their counter-arguments, and some of them are reasonable. Citadel Securities responded to the Citrini scenario with a detailed rebuttal pointing to current labour market data: software engineering job postings rising 11% year on year, new business formation expanding and AI data centre construction creating localised employment booms. Historically, they argue, technological revolutions have altered the composition of tasks rather than eliminated labour as an input. The S-curve of technology adoption, they note, tends to plateau as organisational integration costs mount and diminishing returns set in. This is a fair point. But it is not the whole story.
The argument that new jobs will emerge to replace those automated away is one of the most persistent claims in economic thought. It has a name, though not a flattering one. In Charles Dickens’ David Copperfield, the eternally indebted Mr Micawber lived in cheerful expectation that “something will turn up.” It never did, not until his family emigrated and started again entirely. The Micawber Principle has become shorthand for the belief that optimism, unsupported by any concrete plan, will somehow deliver salvation. When applied to AI and the labour market, this faith that something will turn up, that new tasks and industries will materialise quickly enough to absorb displaced workers, is not economics. It is wishful thinking dressed in a suit.
The historical precedent most often cited by those drawing parallels between AI and earlier technological upheavals is what the economic historian Robert C. Allen termed “Engels’ Pause.” During the British Industrial Revolution, roughly from 1790 to 1840, real wages for the working class stagnated even as output per worker grew significantly. Allen’s 2009 paper in Explorations in Economic History showed that the profit rate doubled during this period while the share of national income going to labour contracted. The surge in inequality was, as Allen put it, intrinsic to the growth process itself: technical change increased the demand for capital, raised profits and squeezed workers for approximately half a century before wages eventually caught up.
Half a century. That is not a brief transitional inconvenience. That is two generations of human suffering during which, as Friedrich Engels himself documented in The Condition of the Working Class in England, urban poverty, child labour and desperate living conditions were the norm, not the exception. The optimists will point out that wages did eventually rise, that the second half of the nineteenth century saw working-class living standards improve substantially. This is true. But as Nobel laureate Daron Acemoglu and Simon Johnson argue in their 2023 book Power and Progress, that improvement was not automatic or inevitable. It required deliberate political choices, the rise of organised labour, regulatory intervention and the redistribution of power away from those who controlled the new technologies. Shared prosperity did not simply turn up. It was fought for and built through institutional reform.
This brings us to the argument for what I would call enlightened self-interest, and it is an argument I would direct particularly at the business leaders and politicians who hold the levers of power in the current AI transition.
Alexis de Tocqueville, writing in Democracy in America in 1835, described what he observed among Americans as “self-interest rightly understood” – the recognition that personal advantage is best served not by short-term exploitation but by sustaining the health of the community in which one operates. As Tocqueville put it, Americans “show with complacency how an enlightened regard for themselves constantly prompts them to assist one another and inclines them willingly to sacrifice a portion of their time and property to the welfare of the state.” The principle, he noted, “checks one personal interest by another, and uses, to direct the passions, the very same instrument that excites them.”
This is not altruism. It is pragmatism. And it applies with particular force to the AI-driven displacement scenario now unfolding.
Consider the economic logic. If AI rapidly displaces a significant proportion of white-collar workers, as the Citrini scenario envisages, the consequences do not remain contained within the labour market. Displaced workers spend less. Reduced consumer spending suppresses demand. Suppressed demand puts further pressure on businesses, which respond with more automation and more layoffs, creating what Citrini described as a negative feedback loop with no natural brake. The Citi note acknowledges this deflationary risk explicitly. If AI’s benefits flow predominantly to capital owners and a narrow technical elite while the broad consumer base is hollowed out, the result is not just social hardship. It is a systemic threat to the very economy from which the elite derive their wealth.
This is why the enlightened self-interest argument should resonate even with those who are unmoved by moral considerations. You cannot sell products to people who have no income. You cannot maintain property values in communities where employment has collapsed. You cannot sustain a financial system built on consumer credit when the consumers are no longer creditworthy. The Citrini report’s concept of “ghost GDP”, economic output that benefits the owners of computing power but never circulates through the human consumer economy, is a vivid way of describing what happens when productivity gains are captured entirely at the top.
I was reflecting on exactly this dynamic during a recent trip to South Korea. Walking through the shopping districts, I was struck by the sheer volume of small shops selling cheap plastic goods, accessories, novelty items and low-margin consumer products. Shop after shop, stacked floor to ceiling, staffed by one or two people, paying rent in cities where commercial property does not come cheap. Even today, it is hard to see how the economics of these businesses add up. The margins on a plastic phone case or a novelty keyring are razor-thin. The rent, the utilities, the wages of the person behind the counter: the sums must be extraordinarily tight. And yet these shops persist, sustained by foot traffic, by impulse purchases, by the simple fact that enough people are walking past with enough disposable income to keep the lights on.
Now project that model into the world the Citrini report describes. In an economy where AI has displaced a significant tranche of white-collar workers, where consumer spending is contracting and deflation is taking hold, those shops do not survive. Not because AI has directly automated what they do, but because the customers walking past their doors no longer have the income or the confidence to spend on non-essential goods. The deflationary cascade does not require AI to enter every shop and every factory. It only requires AI to hollow out enough of the middle-income consumer base for the ripple effects to propagate through every layer of the economy, from financial services to street-level retail. Those small South Korean shops are canaries in the economic coalmine: fragile, low-margin enterprises that depend entirely on a functioning consumer economy. If the consumer economy fractures, they are among the first casualties, and the people who work in them join the growing ranks of the displaced.
This is the point that the techno-optimists consistently underestimate. The threat is not merely that AI automates specific roles. It is that the economic consequences of concentrated displacement create second and third-order effects that reach far beyond the sectors directly affected.
Acemoglu and Pascual Restrepo’s task-based framework, developed across a series of influential papers from 2018 onwards, provides the academic scaffolding for understanding what is at stake. Their work distinguishes between the displacement effect of automation (which reduces labour demand in tasks that machines take over) and the reinstatement effect (the creation of new tasks in which human labour has a comparative advantage). The crucial insight is that there is nothing guaranteed about these two forces remaining in balance. As Acemoglu has warned, the current direction of AI development is weighted too heavily toward automation rather than augmentation. If firms pursue what he calls “so-so automation”, replacing workers with AI that performs only marginally better but at lower cost, the productivity gains will be modest, the displacement will be real and the creation of new tasks will be slow.
The policy implications of this analysis are significant and they deserve serious attention rather than the vague reassurance that markets will self-correct. Acemoglu, Johnson and their collaborators have proposed a number of concrete measures: reforming tax systems that currently incentivise automation over employment; investing in AI applications that complement rather than replace human workers; strengthening worker voice and bargaining power; and – most crucial in my view – ensuring that the direction of technological development is shaped by democratic institutions rather than left entirely to the discretion of a small technology elite.
These are not radical proposals. They are the kinds of institutional adjustments that have historically been necessary during every major technological transition. What is genuinely radical, and dangerous, is the alternative: doing nothing and hoping that the Micawber principle will deliver.
We are at a point where the World Economic Forum projects 92 million workers displaced by AI by 2030. Yes, they also project 170 million new roles created in that period, but these figures mask an enormous transition challenge. The new roles require different skills, exist in different locations and may not materialise at the pace required. The people losing their jobs in financial services, legal support, software engineering and middle management are not seamlessly absorbed into data centre construction or AI safety research. The gap between displacement and reinstatement, which during the Industrial Revolution lasted roughly 50 years, could be just as painful this time around, and the pace of AI development suggests it could arrive far more abruptly.
The case for action is not about being anti-technology. It is about being pro-planning. As Acemoglu and Johnson argue, “there is nothing automatic about new technologies bringing widespread prosperity. Whether they do or not is an economic, social, and political choice.” The elites who benefit most from AI-driven productivity gains have the strongest practical interest in ensuring that those gains are distributed broadly enough to sustain the consumer economy, social stability and democratic institutions on which their own prosperity depends. This is not socialism. It is survival economics.
Those of us working in AI governance have a role to play in shifting this conversation from passive optimism to active planning. We need to help organisations think not just about how to deploy AI efficiently but about how to deploy it responsibly, in ways that augment human capability rather than simply eliminating human roles. We need to advocate for regulatory frameworks that incentivise the creation of new tasks alongside the automation of existing ones. And we need to challenge the comforting narrative that something will turn up.
Because Mr Micawber, for all his charm, spent time in a debtors’ prison. And in the 21st Century, we can’t outrun the economic shift by emigrating to Australia.
