On February 23, markets appeared to momentarily price in the kind of future imagined in Citrini Research’s THE 2028 GLOBAL INTELLIGENCE CRISIS. Shares of International Business Machines(IBM) plunged 13.2% – their steepest daily drop since October 18, 2000. This pluge happened after AI startup Anthropic suggested its Claude Code tool could help modernize legacy COBOL systems running on IBM mainframes.Reuters reported that COBOL, a programming language widely used across banking, insurance and government systems, could be transformed by AI automation. Anthropic wrote: “Modernizing a COBOL system once required armies of consultants spending years mapping workflows. Tools like Claude Code can automate the exploration and analysis phases that consume most of the effort in COBOL modernization.” It added, “With AI, teams can modernize their COBOL codebase in quarters instead of years.”The market reaction was swift. Software and cybersecurity stocks, including CrowdStrike and Datadog, also slid as investors weighed how AI might disrupt established revenue models.A story that spooked the market“The 2028 Global Intelligence Crisis” is a gripping story. It imagines a world, just a couple of years from now, where AI agents hollow out white‑collar work, “Ghost GDP” dominates the national accounts, unemployment blows past 10%, and the consumer economy withers because, as Citrini puts it, machines “spend on discretionary goods. (Hint: it’s zero.)”It’s no accident that a note like this moved markets. As Noah Smith points out in his post titled “The Citrini post is just a scary bedtime story”, a bunch of software and finance stocks sold off right after it went viral. Smith’s read is that this looks less like analysts suddenly discovering a genuine blind spot, and more like a wave of sentiment: traders read an evocative crisis narrative, saw their tickers name‑checked, and panic‑sold in sync.Which jobs and firms get hit?On the micro side, Citrini sketches a world where AI agents “take care of almost all white-collar work, like coding, research, transactions, and even making strategic decisions.” Here the honest answer is that no one knows the exact industry‑by‑industry path. But a few things are clear from current usage. The Economist notes that around 41% of American workers had used generative AI at work by late 2025, yet only 13% of working‑age adults used it daily, and only about 5–6% of work hours involve gen‑AI at all. Most of that is still discrete tasks – drafting, summarising, coding assistance – not fully autonomous agents running entire workflows.When AI is used, task‑level gains are substantial: studies on tools like ChatGPT show completion times for writing tasks falling by nearly 40%, experiments at firms like Boston Consulting Group show 12–25% productivity lifts on realistic professional tasks, and a broader review by academic researchers finds 15–30% gains in real‑world settings.So there is a solid micro story that AI will reshape many white‑collar jobs and business models over time. But that’s a different claim from “this will happen near‑totally and very fast by 2028”.Macro: From disruption to crisis?The stronger claim in Citrini’s piece is macroeconomic: that this rapid micro disruption crashes the whole system. The note imagines an unemployment rate over 10%, a plunge in consumption, and a world where the “human-centric consumer economy, 70% of GDP at the time, withered.”Smith points out that Citrini never really spells out a macro model. They don’t show the transmission mechanism from firm‑level disruption to aggregate crisis; the reader is just told to accept that enough business models break and somehow GDP and employment collapse. He suggests two vaguely plausible channels by which a service‑sector AI productivity boom could end badly: a financial crisis triggered by mass business‑model failure, or a demand shortfall if labour income collapses faster than prices adjust and policy responds.Neither is impossible. But both require strong, specific assumptions: that disruptions arrive faster than balance sheets and regulation can adapt, and that policymakers sit on their hands in the face of a visible technology shock. Historical precedent cuts against that.And the current macro data tell a very different story from the one Citrini is painting. As The Economist stresses, there is little sign of a runaway AI productivity surge in the actual numbers. The US economy grew 2.2% in 2025 while employment basically flatlined, but that translates into productivity growth of roughly 1.9% – just below the post‑war average and far below the internet boom years. Much of that growth came from investment in data centres and related infrastructure. When researchers adjust for that investment, underlying productivity looks “close to zero.”That is not the footprint of an economy already in the grip of an AI super‑cycle that will inevitably blow up by 2028. It is the footprint of an economy that’s still in the build‑out phase, with modest realised productivity gains so far.Ghost GDP, distribution, and policyThe most original concept in Citrini’s essay is “Ghost GDP”: output produced by AI that doesn’t translate into human income, and thus into demand. The worry is intuitive: if AI lets firms do the same work with far fewer people, labour income shrinks, and because “machines spend zero” the system loses its main engine of consumption.Smith’s macro answer is that distribution matters more than aggregates, and policy is not powerless. GDP is just total output; the crucial question is who owns the machines and the profits. Owners are still humans, households, institutions and governments. Their propensity to consume is lower than workers who lose jobs, but it isn’t zero. Dividends, capital gains, and tax receipts all fund spending.
And there’s all this mysterious wealth accumulated by the owners of the GPUs and what are they spending it on? How can there simultaneously be massive wealth and mass layoffs? Will there be new jobs invented? Quite likely. This has been the pattern over the last 200 years: technological revolutions → deflation → demand for new things → new jobs get created. Because humans have infinite desires.
Marcelo P Lima @MarceloLima
History also shows that when job losses mount, governments react. In a world where unemployment genuinely drifts toward 10% because of visible technological change, it is hard to imagine central banks and treasuries shrugging. We have policy tools – automatic stabilisers, direct transfers, wage subsidies, public employment, even variants of basic income or job guarantees – that can turn “Ghost GDP” into purchasing power if voters demand it.None of that guarantees a smooth ride; politics can absolutely misfire. But treating a demand collapse as an unavoidable mechanical outcome of AI, rather than a contingent product of choices, overstates the inevitability of Citrini’s crisis.Why the crisis story still mattersIf “The 2028 Global Intelligence Crisis” is, as Smith puts it, “just a scary bedtime story”, it is still a useful one. It forces people to imagine a world where AI really does rip through white‑collar work and expose how fragile some software‑ and fee‑based business models are; it shows how quickly financial markets can be spooked by narrative alone; and it pushes macroeconomists and policymakers to think about distributional consequences before they arrive.But when we put the story next to the data, and next to a more explicit macro frame, its strongest claims look like outliers. AI will almost certainly take some jobs and transform many more. It might break specific companies and sectors. It might even contribute to a future downturn if shocks and policy mistakes line up in the wrong way.Yet the most likely path for this decade is messier and less cinematic: gradual adoption, uneven productivity gains, choppy labour‑market adjustment, and a rolling policy argument over how to share the benefits. That is plenty challenging without assuming that by 2028 we are living in a world where the machines did great, the economy did not, and the only thing left is Ghost GDP.The story is a warning worth reading. It is not, on current evidence, our baseline future.AI is absolutely going to change who wins, who loses, and how income is shared. It is already rewriting some balance sheets. The evidence so far suggests something harder and less theatrical than a 2028 collapse: a long, uneven slog where productivity gains arrive late, business models are ground down rather than blown up, and politics decides whether “Ghost GDP” stays a metaphor or becomes a statistic.
