From the "June 2028 Research Report": When AI surpasses expectations, the economy crashes

CitriniResearch and Alap Shah’s “Macro Memo from the Future” propose a fictional hypothesis: AI repeatedly surpassing optimistic expectations does not necessarily benefit assets and the economy. On the contrary, abundant machine intelligence may squeeze labor income and consumption cycles, triggering a demand contraction and financial re-pricing driven by a “productivity boom.”

In this thought experiment anchored around June 2028, the U.S. unemployment rate rises to 10.2%, 0.3 percentage points above expectations. After the data release, markets decline 2%, and the S&P 500 has retraced 38% from its October 2026 peak. The memo states that traders have become numb to shocks; six months earlier, similar data might have triggered circuit breakers.

The report breaks down the crisis pathway into two reinforcing chains: one occurring in the real economy, where AI capability improvements displace white-collar jobs, causing real wage growth to collapse, and a “people-centered” economy driven by high consumption begins to shrink, forming a negative feedback loop without natural brakes. Markets initially focus only on AI, but the economy itself starts to deform, giving rise to so-called “Ghost GDP,” where output is recorded in national accounts but fails to circulate in the real economy.

The other chain occurs within the financial system, where structural damage to income expectations begins to erode asset valuations based on white-collar cash flows, such as private credit and mortgage-backed assets, prompting regulatory and policy discussions to accelerate. However, the report emphasizes that policy responses remain delayed, public confidence in government “rescue capabilities” declines, and the risk of a deflationary spiral is amplified.

Perhaps, as Citrini states, “When machine-generated output equals that of 10,000 white-collar workers but consumes not a penny of social services, this is not an economic miracle—it’s an economic plague.”

High profit margins do not equate to a healthy economy: money no longer flows through the household sector

In the scenario, the first wave of layoffs triggered by “human obsolescence” in early 2026 aligns with market preferences: costs fall, profit margins rise, earnings beat expectations, and stock prices climb. By October 2026, the S&P 500 approaches 8,000, and the Nasdaq surpasses 30,000. Corporate profits are funneled back into AI compute power, creating an accelerator effect.

On the macro level, the surface also looks “beautiful”: nominal GDP records multiple high single-digit annualized growth rates, and real output per hour reaches levels the author claims “not seen since the 1950s”—AI agents don’t sleep, don’t call in sick, and don’t need health insurance.

But the memo emphasizes that wealth mainly flows to “owners of compute power,” while labor income collapses. Real wages turn negative, white-collar workers are forced into lower-paid roles, and the “people-centered engine” of consumption—accounting for about 70% of GDP at the time—begins to shrink. The author bluntly asks: How much do machines spend on discretionary consumption? The answer: zero.

SaaS companies are first hit: when “writing your own” becomes a routine procurement option

The first domino in this chain is software. The author sets the inflection point at the end of 2025: a “stepwise leap” in proxy-based programming tools. A competent developer, working with Claude Code or Codex, can replicate core features of a mid-tier SaaS product in a few weeks—imperfect, but enough to make CIOs ask: can we do this ourselves instead of paying $500,000 annually?

Since corporate fiscal budgets are mostly locked in Q4 of the previous year, mid-2026 review becomes the first window for procurement decisions based on “real usable” products. The memo provides a negotiation anecdote: a Fortune 500 procurement manager tells the author he used “discussions with OpenAI about replacing vendors with AI-deployed engineers” as leverage to secure a 30% discount renewal; meanwhile, long-tail SaaS providers like Monday.com, Zapier, Asana face worse conditions.

More critically, how does this “self-build as an option” reshape industry structure? Differentiation is accelerated away by AI development and iteration, turning price wars into “sword fights” against both old competitors and new challengers. Moats are no longer about features but about cost and financing endurance.

Companies threatened by AI are actually the most aggressive: reflexivity cycles begin here

The memo’s most emphasized “non-historical” point is: in 2026, the displaced do not choose to “resist.” Comparing to Kodak, Blockbuster, BlackBerry, the author argues that under AI disruption, many companies “cannot slowly die” but must act swiftly to save themselves.

In this scenario, ServiceNow shows clear signals in Q3 2026: net new ACV growth drops from 23% to 14%, and the company announces 15% layoffs, with its stock falling 18% that day. The reason is straightforward: it sells seat licenses; cut 15% of employees, and 15% of licenses are automatically canceled. The layoffs are driven by AI-driven efficiency gains.

This creates what the memo calls “collective rationality, a systemic disaster”: companies cut costs, then reinvest savings into AI tools, which further enable layoffs. Each company’s actions make sense individually, but collectively, they remove the brakes.

After friction disappears, intermediary layers collapse: from subscriptions, commissions, to card network fees

By early 2027, the author assumes LLMs become the default, with many users “using AI agents as easily as autocomplete,” often unaware. Then, Qwen’s open-source “proxy shopping assistant” acts as a catalyst, with various assistants rapidly integrating proxy e-commerce functions; model distillation allows agents to run on phones and laptops, lowering marginal inference costs.

What unsettles the author most: agents do not need to be explicitly invoked—they run continuously in the background based on preferences. By March 2027, the average American individual consumes about 400,000 tokens daily—10 times more than at the end of 2026. Transactions are no longer a series of discrete human decisions but a continuous optimization process.

This directly undermines the rent layers built on “human limitations” over the past fifty years: automatic subscription renewals, stealth price increases after free trials, brand familiarity replacing diligent price comparison… These friction-based revenue models are transformed into “negotiable hostage situations.”

A list of the earliest “collapse” sectors appears: travel booking platforms, insurance relying on inertia, financial advisors, tax services, routine legal work. Even real estate brokers are not immune: after AI agents gain MLS access and historical transaction data, median buyer agent commissions in major U.S. metros drop from 2.5-3% to below 1%, with more transactions no longer requiring human agents on the buyer side.

Once agents control transactions, they will seek even larger “loops”: in machine-to-machine trading, 2-3% card network interchange fees become glaring. The author envisions many agents settling on stablecoins on Solana or Ethereum Layer 2, with costs approaching “a fraction of a cent.” In this segment, Mastercard is depicted as an “irreversible inflection point”: management mentions “agent-driven price optimization” and “pressure on optional consumption” in earnings calls, leading to stock declines; risks spill over to issuers and single-brand card networks, with AmEx hit hardest (white-collar clients cut + interchange fees bypassed).

This is not about “industry prosperity”: the demand side of the white-collar service economy is being broken by leverage

In 2026, the market still views negative impacts as sector stories—software, consulting, payments. The memo counters directly: the U.S. is a white-collar service economy, with white-collar workers making up about 50% of employment but driving roughly 75% of discretionary consumption.

More pointedly, the memo emphasizes consumption concentration: the top 10% of earners contribute over 50% of total consumption, and the top 20% about 65%. Therefore, if shocks hit high-income white-collar workers, even with relatively modest unemployment, the impact on discretionary spending is disproportionately large. The memo illustrates the leverage: a 2% decline in white-collar employment could correspond to a 3-4% drop in discretionary consumption; and since white-collar workers hold savings buffers, the impact is delayed but potentially deeper.

Specific signals of a turning point in employment are detailed: in October 2026, JOLTS job openings fall below 5.5 million, down 15% YoY; white-collar jobs collapse while blue-collar remains relatively stable. The bond market reacts first: 10-year U.S. Treasury yields fall from 4.3% to 3.2%, reflecting expectations of consumption slowdown.

Meanwhile, AI investment continues unabated despite demand weakness, because the author defines it as “OpEx substitution” rather than traditional CapEx cycles: companies shift $100 million previously spent on labor into AI budgets, reducing total expenditure but increasing AI spending exponentially. This results in a stark divergence: AI infrastructure chains remain high-growth—Nvidia revenues hit new highs, TSMC utilization exceeds 95%, hyperscale cloud providers still spend $150-200 billion quarterly on data centers—while consumer spending begins to hemorrhage.

The author extends this divergence to the national level: South Korea, as a “pure beneficiary,” outperforms significantly; India’s IT services exports (over $200 billion annually) face accelerated contract cancellations as “AI coding proxies’ marginal costs approach electricity prices,” causing the rupee to depreciate 18% against the dollar in four months, with IMF discussions with New Delhi already underway in Q1 2028.

Private credit is not “closed and safe”: life insurance liabilities bring it into the spotlight

The first trigger in the financial layer comes from private credit. The memo details scale changes: private credit grew from less than $1 trillion in 2015 to over $2.5 trillion in 2026, much of it directed toward software and tech LBOs, based on the assumption that SaaS revenues can “compound stably over the long term.”

When AI disrupts ARR sustainability, the problem is not losses per se but the moment losses are recognized. The memo describes several key events: in April 2027, Moody’s downgrades 14 issuers supporting $18 billion of PE-backed software debt all at once; by Q3 2027, software-backed loans begin to default. Zendesk is depicted as a “smoking gun”: its $5 billion direct loan facility supported by ARR is marked down to 58 cents on the dollar, becoming a record-breaking private credit software default.

If it stopped there, the author admits it would be “manageable”—private credit is mostly closed-end, with fixed maturities, so forced runs are unlikely. But “permanent capital” reveals another side: large alternative asset managers, through acquisitions of life insurance companies, have turned annuity liabilities into private credit financing bases (the memo mentions Apollo/Athene, Brookfield/American Equity, KKR/Global Atlantic). As software defaults spread, insurance regulators tighten risk capital requirements, forcing institutions to raise capital or sell assets at unfavorable prices. Moody’s downgrades Athene’s financial strength outlook, leading to a 22% drop in Apollo’s stock over two days, with spillovers to similar firms.

The author adds a more frightening layer of complexity: offshore reinsurance and SPV structures obscure loss attribution, making it difficult to identify who bears the losses. The market crash in November 2027 is depicted as a shift from “cyclical pullback” to “systemic chain reaction”; during an emergency FOMC meeting (within the scenario), Fed Chair Wosh describes it as a “daisy chain of bets on white-collar productivity growth.”

The big risk lies in mortgages: loans were good once, but the world has changed

The memo leaves the “more difficult to price and more deadly” problem to housing mortgages. The U.S. residential mortgage market is about $13 trillion, underwritten on the assumption that borrowers’ employment and income remain stable over long periods (often 30 years).

The scenario’s key risk: this is not a 2008-style “bad loans from the start.” Instead, borrowers are “model citizens”: FICO scores above 780, 20% down payments, verifiable income, clean credit histories. The problem is that, after structural downgrades in white-collar income expectations, the future cash flows underpinning these “pillars” are no longer credible—people are borrowing against a future they increasingly doubt.

The memo lists early warning signs: HELOC drawdowns, 401(k) early withdrawals, rising credit card debt, yet mortgage payments remain current; then, delinquencies begin to rise in places like San Francisco, Seattle, Manhattan, Austin. By June 2028, Zillow house price indices show YoY declines: San Francisco -11%, Seattle -9%, Austin -8%; Fannie Mae reports that high-end (jumbo) loans in zip codes with over 40% tech/financial employment are experiencing higher early delinquencies.

The author deliberately keeps the boundary: the “full-blown mortgage crisis” has not yet arrived in this scenario; delinquency levels are still well below 2008, but the trajectory is clear. If mortgages truly crack in the second half of that year, the author estimates stock market retracement could approach 57%—similar to the 2008 global financial crisis—and the S&P could fall toward 3,500 points—close to the “ChatGPT moment” of November 2022.

Policy’s greatest enemy is time: tax bases are built on human time

The memo is blunt about policy: traditional tools (rate cuts, QE) can save the financial engine but struggle to repair the real economy, because the root cause is not “money too expensive” but “human intelligence becoming less valuable.”

More specific fiscal constraints are summarized in one sentence: federal revenue is essentially a tax on human time—people working, companies paying wages, government collecting. By Q1 2028, federal receipts are 12% below CBO baseline. Productivity soars, but the gains flow more to capital and compute ownership, not back to households via income or payroll taxes.

Long-term decline in labor share of GDP is the backdrop: from 64% in 1974 to 56% in 2024; after four years of AI-driven exponential improvement, it further drops to 46%, the steepest recorded decline, the author says.

This creates a structural paradox: the government needs to transfer more funds to households but collects less in taxes. The scenario discusses proposals like the “Transition Economy Act” (funded by deficits and a tax on AI compute) and the more aggressive “Shared AI Prosperity Act” (establishing a public claim on “intelligent infrastructure returns,” akin to sovereign wealth funds or AI output royalties, with dividends supporting transfers). Political divisions are sharp: right-wing critics call transfers “Marxist,” worry about the dominance of Chinese compute taxes; left-wing critics fear regulatory capture; fiscal hawks warn that deficits are unsustainable, while doves cite the premature tightening after the GFC as a cautionary tale.

Societal friction surfaces: in the scenario, “Occupy Silicon Valley” protests block Anthropic and OpenAI offices in San Francisco for three weeks, drawing media attention even more than unemployment data. The conclusion is that institutional change cannot keep pace with technological change, and feedback loops will drive political decisions.

“Intellectual premium” unwinds: the need to re-evaluate cash flow assumptions from the old world

The memo attributes all this to a deeper shift in pricing: in modern economic history, human intelligence has been a scarce factor—central to labor markets, mortgage underwriting, tax systems, and corporate moats. Now, as machine intelligence becomes a substitutable and increasingly cheap resource, the “intellectual premium” begins to unwind, forcing the financial system to reprice painfully.

The author leaves room for hope: re-pricing does not necessarily mean collapse; the economy might find a new equilibrium. The challenge is whether it can do so before the feedback loops write the next chapter. As of February 2026, the S&P remains high, and negative feedback has not yet started. The author’s warning is more like a self-check for investors: how much of your assets and cash flows are still based on the assumptions that “frictions will not disappear, white-collar income will stay stable, households will continue to drive demand”? The final line echoes the canary: the canary is still alive.