Why AI Disrupts Demand Faster Than Supply
Post-Keynesian underconsumption theory finds its clearest test case yet in agentic AI.
The Citrini Research post landed like a grenade in a room full of optimists already half-nervous about AI. Published February 22, 2026, it sketched an intelligence displacement spiral that turned runaway AI success into economic suicide. Agentic systems would replicate SaaS tools overnight, gut white-collar headcounts, spike corporate margins briefly, then choke demand as high-earning consumers vanished from the spending pool.
Machines produce. Humans stop buying. Velocity craters.
The piece framed itself as a pre-mortem thought exercise, not a forecast. Markets treated it as a credible catalyst anyway.
Monday’s trading session delivered the proof. IBM cratered 13 percent, its worst single-day wipeout since 2000, tangled partly with fresh Anthropic announcements on Claude modernizing legacy COBOL but amplified by the Citrini narrative tagging legacy enterprise software as prime displacement fodder. DoorDash, Uber, American Express, Mastercard, Visa, KKR, Blackstone: all slid four to eight percent or worse. Bloomberg pinned the move squarely on the viral Substack post by James van Geelen and Alap Shah. What started as a hypothetical tail risk crystallized existing jitters into a coordinated selloff across anything touched by intermediation, payments, or delivery.
The mechanism Citrini outlined mirrors the demand-collapse logic Post-Keynesians and Michal Kalecki hammered for decades. Efficiency gains that slash labor costs look brilliant on quarterly calls. But when the top earners who power most discretionary consumption lose jobs or income, the consumption engine stalls. Say’s Law collapses under its own weight. Supply creates its own demand only if incomes stay distributed roughly as before. Here, income shifts violently toward capital owners and AI infrastructure, who save or reinvest rather than spend on DoorDash runs or vacations. National statistics report growth; the real economy contracts.
Wall Street’s reflexive cheer for lower costs equals higher margins exposed the same orthodox blind spot Kalecki diagnosed in the 1930s. Markets framed the initial AI surge as unambiguously bullish until the Citrini scenario forced acknowledgment that unchecked labor substitution could trigger underconsumption on a scale past automation waves never reached. Previous tech shifts displaced some roles but spawned net new demand and occupations. Citrini argues agentic AI hits different: it attacks cognitive friction across the board, from coding to insurance shopping to payment routing via stablecoins that sidestep interchange fees. No offsetting job boom materializes fast enough.
Citrini proposed remedies that lean liberal-redistributive. A Transition Economy Act would channel deficit spending into direct transfers for displaced workers, funded partly by taxing AI inference compute. The more ambitious Shared AI Prosperity Act envisions public royalties on intelligence infrastructure, creating a sovereign-style fund that pays household dividends. These steps recognize that sovereign currency issuers face no solvency constraint and can deficit-spend without default risk. Yet they remain largely passive: money moves from AI winners to displaced workers via transfers, stabilizing demand but not necessarily rebuilding robust human-centric multipliers.
Kalecki offered a sharper alternative. In his 1943 essay on full employment politics, government should directly create jobs through public investment, schools, hospitals, infrastructure, or subsidize mass consumption via allowances and price supports. The engine is active employment policy, not just checks. Kalecki also stressed the political obstacle: businesses resist full employment because it shrinks the reserve army of labor, strengthens unions, and threatens capitalist control over investment direction. They prefer sound finance sermons while quietly benefiting from demand stimulus they publicly decry.
The divergence matters. Citrini’s toolkit cushions the spiral through redistribution; it may blunt the worst but leaves the human economy hollowed. Kalecki’s path attacks the root by restoring wage-earner purchasing power through work. Both see the crisis as structural and both dismiss monetary tools alone as plumbing fixes for a broken engine.
Skeptics like Noah Smith push back hard. AI will spawn new income streams, deflate prices to boost real purchasing power, and meet adaptive policy responses. The Citrini scenario, vivid as it is, risks over-indexing on left-tail fears while underplaying adaptation.
Still. When a hypothetical memo from 2028 can erase hundreds of billions in a single session, the demand-side blind spot stands exposed once more. Eighty years after Kalecki mapped these traps, the lessons refuse to stay buried. AI only sharpens the blade.