Last week CitriniResearch published The 2028 Global Intelligence Crisis, co-authored with Alap Shah of LOTUS — a speculative macro memo written from the perspective of June 2028, looking back at how AI broke the economy. Shah posed the original question, and together they built a sharp, well-researched scenario. (Shah has also written companion pieces in his Intelligence Explosion series worth reading alongside this one.)
The thesis: AI works exactly as well as the bulls expect, companies replace white-collar workers, those workers stop spending, the consumer economy withers, and a negative feedback loop spirals into a financial crisis. They call it the “Intelligence Displacement Spiral” — a loop with no natural brake.
It’s good work. I want to build on it, because I think there’s a missing half that changes the picture.
What Citrini and Shah Got Right
Go read their piece first — this post is a response, not a substitute. They identify real mechanisms:
The negative feedback loop is sound. AI capability improves, companies cut headcount, savings get reinvested in AI, displaced workers spend less, weaker demand pushes more companies to cut. Each company’s decision is individually rational. The collective result is destructive.
The financial contagion chain is real. Private credit loaded with PE-backed SaaS deals, funded by insurance company annuity deposits, linked through offshore reinsurance structures. When the underlying loans go bad, the opacity of who holds the losses is dangerous. They trace the chain from Zendesk’s $10.2B LBO defaulting through to Athene’s financial strength rating getting downgraded, and the mechanics hold up.
The mortgage insight is sharp. In 2008, the loans were bad on day one. In their 2028, the loans were good on day one — the world just changed after they were written. A 780-FICO product manager who put 20% down isn’t subprime. But if their job category is being structurally eliminated, the income assumptions underlying the mortgage are still broken.
“Ghost GDP” is a useful concept. Output that shows up in national accounts but never circulates through the real economy. A GPU cluster in North Dakota does the work of 10,000 Manhattan workers. It produces GDP. It doesn’t produce restaurant visits, apartment leases, or income tax revenue.
All worth taking seriously. But the piece rests on a framing choice that limits its conclusions.
The Framing Problem
The article models AI exclusively as labor substitution — the same economy, minus humans. Every section is about replacing workers, eliminating friction, and compressing margins on existing activity. It never models AI as a capability multiplier that creates things humans couldn’t do before.
That matters. There are three big counterweights missing from their scenario.
1. The Physical World Isn’t Software
The article treats the economy as frozen in its current shape, with AI just removing humans from it. But AI plus robotics doesn’t just make existing factories cheaper — it makes new categories of manufacturing possible.
Generative design is producing parts no human engineer would conceive of. AI-optimized supply chains can run lights-out manufacturing. Small-batch custom production that was never economically viable is becoming routine.
Their own scenario has all of GE Vernova’s turbine capacity sold out through 2040. They mention it as evidence that AI infrastructure is booming — but never ask who’s building all that capacity, who’s installing it, who’s maintaining it, and what jobs that creates.
The physical world is a bottleneck that AI alone can’t solve. You still need people to install, maintain, permit, inspect, and operate. Electricians for data centers. Technicians for robotics lines. Construction workers for the factories. The article skips past all of this.
2. Science Breakthroughs Create New Economies
This is the biggest gap. The article never mentions:
Drug discovery. AI is compressing 10-year, $2 billion pipelines into months. Antibiotics, cancer treatments, rare disease therapies that were never economically viable to develop are entering trials. Each one creates manufacturing, distribution, clinical, and care delivery work.
Materials science. Novel battery chemistries, advanced alloys, carbon capture materials, next-gen semiconductors. These create new industries. Every breakthrough material needs production facilities, quality control, application engineering, and installation.
Agriculture. Precision farming, drought-resistant crop design, synthetic biology. AI-accelerated agricultural science is creating markets that don’t exist today.
Energy. Fusion research accelerated by AI, next-gen solar, grid optimization. The energy transition is a multi-trillion-dollar buildout that needs workers for decades.
The article treats the economy as a fixed pie where AI shifts who gets the slices. AI applied to science grows the pie — new goods, new treatments, new capabilities that generate new demand.
They argue “this time is different because AI can do the new jobs too.” That’s true for information work. AI can’t pour a foundation, administer a gene therapy, retrofit a building for new materials, or run a clinical trial on humans. Science breakthroughs create demand for physical-world work.
3. The Demographic Cliff
This one potentially inverts the thesis.
The article’s scenario: too many workers, not enough jobs. The actual macro trend heading into the 2030s is the opposite — the developed world is running out of workers.
- US fertility rate: 1.62 (replacement is 2.1)
- China’s population is already declining
- Japan and South Korea are well below replacement
- Most of Europe is too
- The dependency ratio (retirees per working-age adult) is heading toward crisis levels
Social Security and Medicare were designed for 3+ workers per retiree. We’re heading toward 2:1. Without productivity gains, the 2030s were shaping up to be: not enough humans to care for aging populations, staff hospitals, maintain infrastructure, or sustain output.
AI arriving as the demographic cliff hits isn’t a labor catastrophe. It might be the only thing that prevents one.
The “displacement” the article fears may be filling roles that were going to be unfillable. Japan is the preview — they’re not worried about AI taking jobs. They need it because they don’t have enough people.
The article models a world with static labor supply and declining labor demand. The actual world has a declining labor supply that AI may be arriving just in time to offset.
The Transition Is Still Real
None of this makes the article’s warnings irrelevant. The transition risk is real even if the destination is good.
Speed mismatch. AI capabilities move quarterly. Retraining, new industries, and policy move in years. People caught in the gap suffer regardless of the long-term outlook.
Distribution. Productivity gains flow to capital owners first. Without policy intervention, the benefits concentrate just as badly as the article predicts.
Financial contagion doesn’t care about the long term. Markets can panic on a 2-year timeline even if the 10-year outlook is fine. The private credit chain and mortgage market are vulnerable to a transition that moves faster than institutions can adapt.
Not everyone transitions. A 55-year-old product manager isn’t becoming a gene therapy technician. Real people need real support, and the policy infrastructure barely exists.
A Better Framework
The article models: AI = same economy minus humans = collapse.
More completely: AI = different economy + demographic offset + new capabilities + transition pain.
The better historical analogy probably isn’t the Great Depression. It’s the post-WWII transition — massive restructuring, new capability coming online, real pain for specific groups, but a more productive economy on the other side.
The difference: that transition had the GI Bill, the Marshall Plan, and massive infrastructure spending to bridge the gap. We don’t have the policy equivalent yet. On that point, Citrini and Shah are right to push for urgency.
The Real Question
The article ends with: “The canary is still alive.”
I’d add: the song is more complex than a single note. AI is both the most deflationary force in the economy and the most powerful capability creator we’ve built. Both are true. Citrini and Shah wrote the deflationary half well. But you can’t model abundant intelligence without also modeling what it creates — not just what it replaces.
Whether the next decade is a crisis or a transformation depends on whether we build the bridges fast enough. The technology isn’t the constraint. Policy, retraining, and distribution are.
Their scenario is worth reading. Go read it. Then think about the other half.