We all worry about being replaced by AI, but what did Citrini's apocalyptic prediction overlook?

Excellent articles can cause the market to confuse “scenario planning” with “prophecy.”

On February 22, 2026, a report titled “The 2028 Global Intelligence Crisis” ignited social media and financial markets, with over 27 million views. On the day of its release, IBM plummeted 13%, and stocks of DoorDash, American Express, KKR, and others fell more than 6%.

The report was authored by James van Geelen, founder of Citrini Research. This 33-year-old researcher has over 180,000 followers on X, and his Substack ranks first among finance writers, focusing on equity investment and global macro research. His style is known for cross-asset and lateral associations, with a real investment portfolio returning over 200% since 2023. The report uses scenario planning to imagine a future set in 2028: AI rapidly replacing white-collar workers within two years, leading to consumption contraction, software asset defaults, credit tightening, and ultimately pushing the economy into a distorted state of “technological prosperity” and “social decline.” Van Geelen notes at the beginning: “This article discusses a possible scenario, not a prophecy.” But the market clearly has no patience to distinguish between the two.

However, more than the brief market panic, what’s worth paying attention to is the widespread discussion this article has sparked over the past few days. From academia to investment circles, from Wall Street to the Chinese internet, responses from various perspectives have emerged. Instead of blindly trusting one extreme conclusion, perhaps we can piece together a clearer future from the “disagreements and overlaps” among different viewpoints.

What Citrini Said

The logic in Citrini’s article isn’t complicated: rapid advances in AI capabilities lead to large-scale replacement of white-collar jobs → rising unemployment causes consumption to shrink → structured financial products based on SaaS assets face defaults → credit tightens across the broader financial system → the economy falls into a distorted state of “technological prosperity” and “social decline.”

Each link in this causal chain is supported by evidence. An empirical study by Bick, Blandin, and Deming, based on corporate expenditure data, shows that after ChatGPT’s release, companies with the highest AI exposure (those that previously spent the most on online labor markets) significantly increased their spending on AI model providers while reducing online labor market expenditures, with a decrease of about 15%. Notably, this substitution isn’t “one-to-one”—for every dollar cut from labor market spending, companies only increased AI spending by about $0.03 to $0.30. In other words, AI is accomplishing the same work at a fraction of the cost of human labor.

But Citrini may overestimate the speed of this transition. Critics cite the U.S. real estate agent industry as an example: despite the technology’s ability to drastically reduce the number of agents, the industry still employs over 1.5 million people. Institutional inertia, regulatory barriers, and internal industry interests form a much stronger barrier than technology alone. They argue that Citrini severely underestimates the resistance posed by “institutional momentum.”

Others cite research by Kimball, Basu, and Fernald (1998), which indicates that technological shocks historically tend to be positive supply-side stimuli—short-term employment adjustments may occur, but long-term output increases far outweigh job destruction.

In fact, historically, each wave of general-purpose technology takes much longer to diffuse from labs to widespread adoption than the technology’s own maturity. It took 30 years for electricity to reach 50% household penetration, 35 years for telephones, and even the fastest-growing smartphones still took about 5 years. While AI’s capabilities may already threaten many industries, the gap between technological potential and institutional capacity to absorb it has never been bridged solely by technological ability.

The second key link in Citrini’s narrative is the demand-side spiral: unemployment → income reduction → consumption contraction → corporate profits decline → further layoffs.

Citrini confuses demand-side deflation with supply-side deflation here. The former means consumers’ purchasing power shrinks; the latter refers to technological progress lowering production costs—AI-driven price declines are more akin to the latter, similar to the trajectory of electronics and communication services over the past decades. Some analysts believe that the Jevons paradox will still apply: as AI drastically reduces costs for legal consulting, medical diagnostics, software development, and other services, demand previously excluded by high prices will be unleashed, leading to explosive growth rather than contraction. Meanwhile, the “Moravec paradox” also plays a role: for machines, the real challenge is not complex reasoning or data retrieval but human-like physical movements, sensory perception, and emotional communication. This suggests that physical labor and finely perceptive service jobs may be more resilient than we think.

However, Jevons’ paradox could also fail. Alex Imas, an economist at the University of Chicago, argues that if AI automates most labor and labor’s share of total income sharply declines, who will buy the goods and services produced? This touches on distribution mechanisms. When output capacity approaches infinity and effective demand becomes concentrated, we may face not a recession but an imbalance—material abundance but inaccessible to most.

Glimpsing the Big Picture

The most expansive part of Citrini’s scenario is the transmission from employment shocks to a financial crisis. He envisions structured financial products (“Software-Backed Securities”) based on SaaS revenues experiencing widespread defaults amid the AI transition, triggering a credit crunch similar to 2008.

However, critics point out that compared to 2008, the leverage of U.S. corporations today is much healthier, and the banking system is far more resilient after Dodd-Frank reforms and stress tests.

Compared to the pre-2008 financial crisis, U.S. financial resilience indicators have improved significantly: Tier 1 capital adequacy ratio increased from 8.1% to 13.7%, household debt-to-disposable income ratio decreased from 130% to 97%, and non-performing loan rates fell from 1.4% to 0.7%.

Even if some SaaS companies face revenue declines, their scale is insufficient to trigger systemic credit crises. Nick Smith, a former Bloomberg finance columnist, argues that Citrini makes a common mistake: extrapolating micro-level industry shocks linearly to macro-level systemic risks. Regarding demand collapse, Smith suggests fiscal policy as the solution. If unemployment truly rises sharply, governments have the capacity and willingness to implement large-scale fiscal stimulus to support demand.

The ability of institutions to respond is also underestimated. During COVID-19, for example, after WHO declared a pandemic on March 11, 2020, the U.S. enacted the $2.2 trillion CARES Act just 16 days later. Over the following year, the U.S. deployed a total of $5.68 trillion in fiscal stimulus, about 25% of 2020 GDP.

If AI-driven unemployment occurs at the speed and scale Citrini describes, policy interventions are unlikely to be absent.

Some critics also raise fundamental questions. The doomsday view of technology often stems from a lack of faith in human institutions. Citrini’s scenario treats markets as a machine operating without human oversight, driven solely by causal chains until collapse. But in reality, economic systems are deeply influenced by law, institutions, politics, culture, and ideology.

Consensus and Disagreement

We might attempt to identify some points of consensus and disagreement.

Almost everyone agrees that AI is, and will continue to be, changing the demand structure for white-collar labor. The debate is about the speed and scale of this change. Additionally, the pain of transition is real and should not be masked by long-term optimism. Moreover, the quality and speed of policy responses will largely determine the outcome.

Disagreements lie in the underlying logic. Some believe this wave of technological impact may surpass historical precedents in speed and scope, limiting the applicability of historical analogy; others trust in the adaptability of institutions and the repeatability of history.

Looking Ahead

Citrini’s article has several issues: overly tight logical connections, underestimation of institutional responses, and a leap from micro-industry shocks to macro systemic risks lacking sufficient intermediate reasoning. But its fundamental flaw may be underestimating human society: it assumes a static institutional environment where technology relentlessly and almost irresistibly crushes everything. Throughout history, doomsday scenarios for technology have been numerous, often logically impeccable, yet almost invariably neglect the variable “human.” The complexity, friction, redundancy, and seemingly inefficient institutional arrangements of human society form a powerful, distributed resilience. We have ample time to avoid the apocalyptic outcomes depicted, provided we are not intimidated by the scenarios themselves.

What about optimistic narratives? The Jevons paradox is an observation about long-term trends. The Moravec paradox tells us that physical labor is temporarily safe but does not address what happens to displaced white-collar workers. Historical analogy is insightful but never exact; it merely provides a rhythm. Optimistic stories require time to test, and we are at the starting point of that test.

Doomsday narratives produce anxiety, and those anxious pay the price. Developing your own judgment, bearing risks, managing positions—these are the ways forward, rather than drowning in articles that seem to see only the end.