In February 2026, two serious analysts published contradictory takes on AI and labor displacement. Citrini Research wrote a speculative scenario — framed as a June 2028 retrospective memo — describing how AI-driven displacement could cascade into a structural economic crisis. Frank Flight at Citadel Securities published a point-by-point empirical rebuttal almost immediately.
Both pieces are worth reading. Neither is obviously wrong. The disagreement is real, and understanding where they actually diverge is more useful than picking a side.
What Each Side Is Arguing
Citrini’s piece is scenario planning, not a forecast. They explicitly say so. The exercise is: if AI capabilities continue advancing along their current trajectory, what would a plausible left-tail outcome look like? The scenario walks through four stages — enterprise software collapse, consumer-facing disruption, a displacement spiral, and a mortgage/credit cascade — playing out over roughly two years.
Flight’s rebuttal attacks the scenario as if it were a prediction. His core argument: current data shows stable unemployment (4.28%), rising software job postings (+11% YoY), and AI adoption curves that track slower than previous technologies. The evidence for imminent catastrophe simply isn’t there.
That framing mismatch matters. Citrini never claims 2028 is certain. Flight never claims AI is harmless forever. They’re partly talking past each other — but where they’re talking directly to each other, the disagreements are substantive.
Where They Genuinely Disagree
Is there a natural brake?
This is the crux.
Flight’s argument: Compute economics create a natural ceiling. As AI substitution accelerates, compute demand rises, pushing marginal costs upward. At some point, compute costs exceed labor costs for specific tasks — and substitution becomes economically irrational. The economy self-corrects.
Citrini’s counter: The cause of displacement cannot be its own corrective. Unlike inventory overshoot (which reverses when restocking begins) or overbuilding (which reverses when rates drop), AI improving and getting cheaper is not a mechanism that reverses. Rate cuts to zero don’t change the fact that an AI agent costs $200/month while a mid-level knowledge worker costs $180,000/year. Flight’s ceiling may exist in theory, but it’s a long way off and the cost curves are moving fast in the wrong direction.
This is an unresolved empirical question. Flight is probably right that there’s some ceiling. Citrini is probably right that we’re not close to it yet.
Is this tech wave categorically different?
Flight leans on history: steam, electricity, computing — every major technology followed an S-curve and expanded consumption rather than destroying it. Keynes predicted 15-hour workweeks in 1930; instead, societies consumed more. Productivity shocks are disinflationary and growth-enhancing, not demand-destroying.
Citrini argues the analogy fails: “This is the first time in history the most productive asset in the economy has produced fewer, not more, jobs.” Previous waves replaced physical labor or narrow cognitive tasks. The redeployment question then was: where can displaced workers go? When steam replaced manual labor, workers moved into cognitive work. When AI is a general-purpose intelligence, redeployment into what?
Citrini’s structural argument here is harder to dismiss than Flight acknowledges.
Which workers get hit — and why it matters
Flight treats displacement as a general labor phenomenon and applies the standard framework: productivity expands aggregate demand, everyone eventually benefits.
Citrini targets the specific mechanism: this shock hits the top income decile first — the 10% of earners who account for roughly 50% of discretionary spending. That’s not the factory-worker displacement of past waves. If the highest earners take the first hit, the demand consequences are qualitatively different.
There’s also an embedded assumption in Flight’s model that Citrini keeps pressing: productivity gains must circulate through wages to expand aggregate demand. If AI productivity accrues almost entirely to compute owners — what Citrini calls “Ghost GDP” — the accounting identity holds but the demand stimulus doesn’t materialize. Flight never engages with this. He assumes gains circulate. Citrini’s entire scenario depends on them not circulating fast enough.
The private credit question
Flight focuses on public macro indicators: unemployment, job postings, adoption curves. They all look stable.
Citrini argues the fragility is already embedded in private markets, invisible in public data:
- PE-backed SaaS companies still valued at acquisition multiples from 2021–2022, while comparable public companies trade at 50 cents on the dollar
- ARR-backed loans structured on the assumption that software revenue stays recurring — Zendesk’s $10.2B LBO being the marquee example
- Life insurers with structural exposure to private credit through Apollo/Athene-style alt-manager arrangements
Flight doesn’t address any of this. It’s the most significant gap in his rebuttal.
The Scorecard
Flight wins on:
- Current empirical data. Unemployment is stable. Software job postings are rising. Adoption is following S-curves, not hockey sticks.
- The compute ceiling argument is underrated. It’s a real constraint.
- Correctly notes that displacement fears have been premature in every prior tech wave.
Citrini wins on:
- The structural argument that general intelligence is categorically different from domain-specific automation.
- The “top decile displacement” demand mechanism — more precise than Flight’s aggregate demand framing.
- The private credit / PE cascade fragility — a genuine vulnerability Flight never addresses.
- The “no natural brake” framing — hard to counter if agentic AI keeps improving and costs keep falling.
- The “productivity gains must circulate” assumption doing heavy lifting in Flight’s rebuttal without examination.
The Honest Read
Flight is probably right about timing — 2028 is likely too soon. Citrini is probably right about direction — this is structurally different from past tech waves and warrants serious scenario planning.
The unresolved question at the center of this debate: does the classic “productivity shocks expand consumption” principle hold when productivity gains accrue almost entirely to compute owners rather than distributing through wages? That assumption is doing enormous work in Flight’s rebuttal. Citrini never lets it stand unchallenged.
If gains circulate — Flight is right. If they don’t — Citrini’s scenario becomes far more plausible than Flight’s historical analogies suggest.
Sources: Citrini Research — The 2028 Global Intelligence Crisis; Frank Flight / Citadel Securities — The 2026 Global Intelligence Crisis