
2028 Global Intelligence Crisis
Original: THE 2028 GLOBAL INTELLIGENCE CRISIS Author: CitriniResearch · February 22, 2026
What if our bullish thesis on AI has been right… but that’s actually bearish?
This is a scenario exercise, not a prediction. It’s not short porn, nor is it AI doomer fan fiction. The sole purpose of this article is to model a scenario that has been under-discussed so far. Our friend Alap Shah raised the question, and we brainstormed the answer together. We wrote this part; he wrote two others, which can be found here.
Hopefully, after reading this, you’ll be better prepared for the left-tail risks that could emerge as AI makes the economy increasingly strange.
This is CitriniResearch’s June 2028 macro memo, detailing the evolution and consequences of the Global Intelligence Crisis.
📷
CitriniResearch February 22, 2026 / June 30, 2028
Unemployment came in at 10.2% this morning, 0.3% above expectations. Markets dropped 2% on the news, bringing the cumulative drawdown from the October 2026 highs to 38% on the S&P 500.
Traders are numb. Six months ago, a number like this would have triggered a circuit breaker.
Two years. From “manageable” and “sector-specific” to an economy unrecognizable, in just two years. This quarter’s macro memo is our attempt to reconstruct the process—a post-mortem of the pre-crisis economy.
The Party
By October 2026, the S&P 500 was approaching 8,000 and the Nasdaq had broken 30,000. The first wave of layoffs driven by human obsolescence began in early 2026, and they had exactly the effect layoffs are supposed to have: margin expansion, earnings beats, stock price increases. Record corporate profits were recycled directly into AI compute investment.
Headline data still looked strong. Nominal GDP was repeatedly posting mid-to-high single-digit annualized growth. Productivity was surging. Real output per hour was growing at a pace not seen since the 1950s, driven by AI agents that don’t sleep, don’t call in sick, and don’t need health insurance.
Compute owners saw explosive wealth growth as labor costs vanished. Meanwhile, real wage growth collapsed. Despite the government’s repeated boasts about record productivity, white-collar workers were being displaced by machines and forced into lower-paying jobs.
As cracks appeared in the consumer economy, economic commentators coined a term—"ghost GDP": output recorded in national accounts that never circulates in the real economy.
AI was exceeding expectations in every way, and the market was AI. The only problem was… the economy wasn’t.
We should have seen it coming: a GPU cluster in North Dakota producing the equivalent output of ten thousand white-collar workers in midtown Manhattan—this was less an economic tonic and more an economic pandemic. The velocity of money was grinding to a halt. The human-centric consumer economy (still 70% of GDP at the time) was shrinking. If we had asked earlier, “How much do machines spend on discretionary consumer goods?” we might have figured it out sooner. (Hint: the answer is zero.)
AI capability improves → companies need fewer employees → white-collar layoffs increase → laid-off workers reduce consumption → margin pressure forces companies to invest more in AI → AI capability improves further…
A negative feedback loop with no natural brake. The human intelligence substitution spiral. The earning power of white-collar workers (and, rationally, their consumption) suffered structural damage. Their income was the foundation of the $13 trillion mortgage market—forcing underwriters to reassess whether prime mortgages were still safe.
Seventeen years without a real default cycle, and the private markets were filled with PE-backed software trades assuming ARR would keep growing. The first wave of defaults triggered by AI disruption in mid-2027 shattered that assumption.
If the disruption had been limited to software, it might have been manageable. But it wasn’t. By late 2027, it threatened every business model whose profit came from intermediation. Companies that had built fortunes by creating friction for humans were collapsing.
The entire system turned out to be a correlated domino chain of bets on white-collar productivity growth. The November 2027 crash only accelerated every negative feedback loop already in place.
We’ve been waiting for “bad news is good news” for nearly a year. The government has started considering plans, but public confidence that the government can implement any rescue has faded. Policy responses always lag economic reality, but the lack of a comprehensive plan is threatening to accelerate the deflationary spiral.
The Collapse of SaaS
In late 2025, agentic programming tools achieved a step-function improvement in capability.
A reasonably skilled developer using Claude Code or Codex could replicate the core functionality of a mid-tier SaaS product in a few weeks. Not perfect, not handling every edge case, but enough to make a CIO reviewing a $500,000 annual renewal start asking: “What if we build it ourselves?”
Fiscal years mostly align with calendar years, so 2026 enterprise spending was locked in during Q4 2025, when “agentic AI” was still a buzzword. Mid-year reviews were the first time procurement teams made decisions with actual knowledge of these systems’ capabilities. Some watched their own internal teams build prototypes in weeks that could replace six-figure SaaS contracts.
That summer, we spoke with a procurement manager at a Fortune 500 company. He described a budget negotiation. The salesperson expected to repeat last year’s script: 5% annual price increase, standard “your team can’t live without us” pitch. The procurement manager told him he’d been talking to OpenAI about having their “frontline deployment engineers” use AI tools to completely replace this vendor. The final deal was a 30% discount renewal. He said that was a good outcome. The “long tail” of SaaS, like Monday.com, Zapier, and Asana, was in much worse shape.
Investors were prepared—even expecting—the long tail to be hit hard. It might account for a third of typical enterprise tech stack spending, but it was clearly exposed. Core systems, however, were supposed to be immune to disruption.
Until ServiceNow’s Q3 2026 earnings, the reflexivity mechanism became clearer.
SERVICENOW NET NEW ACV GROWTH SLOWED FROM 23% TO 14%; ANNOUNCED 15% LAYOFFS AND “STRUCTURAL EFFICIENCY PROGRAM”; SHARES FELL 18% | Bloomberg, October 2026
SaaS didn’t “die.” There was still a cost-benefit trade-off for building and maintaining in-house. But building became an option, and that affected pricing negotiations. Perhaps more importantly, the competitive landscape had changed. AI made it easier to develop and ship new features, so differentiation collapsed. Incumbents were caught in a price war—both with each other and with new challengers rising on the back of agentic programming capabilities, unburdened by legacy cost structures, aggressively taking market share.
The interconnected nature of these systems wasn’t fully appreciated until that earnings report. ServiceNow sold seats. When Fortune 500 customers laid off 15% of their workforce, they canceled 15% of their licenses. The AI-driven layoffs that boosted customer margins were mechanically destroying ServiceNow’s own revenue base.
A company selling workflow automation was being disrupted by better workflow automation, and its response was to lay off staff and invest the savings in the technology disrupting it.
What else could they do? Sit back and slowly die? The companies most threatened by AI became the most aggressive adopters of AI.
In hindsight, this seems obvious, but at the time it really wasn’t (at least to me). Historical disruption models said: incumbents resist new technology, lose to agile new entrants, and slowly die. Kodak, Blockbuster, BlackBerry—all of them. What happened in 2026 was different—incumbents didn’t resist because they couldn’t afford to.
Stock prices fell 40-60%, boards demanded explanations, and companies threatened by AI did the only thing they could: lay off staff, reinvest the savings into AI tools, and use those tools to maintain output at lower cost.
Each company’s individual response was rational. The collective result was catastrophic. Every dollar saved in labor costs flowed into AI capabilities that enabled the next round of layoffs.
Software was just the opening act. While investors were still debating whether SaaS valuations had bottomed, the reflexivity loop had already escaped the software industry. The same logic that justified ServiceNow’s layoffs applied to every company with a white-collar cost structure.
The Death of Intermediation
By early 2027, LLM usage had become the default. People were using AI agents without knowing what an AI agent was, just like people who never understood “cloud computing” still used streaming services. They saw it as something like autocomplete or spell check—a new feature on their phone.
Qwen’s open-source agentic shopping assistant became the catalyst for AI taking over consumer decision-making. Within weeks, every major AI assistant had integrated some form of agentic commerce. Distilled models meant these agents could run on phones and laptops, not just in the cloud, dramatically reducing the marginal cost of inference.
What should have made investors more uneasy was that these agents didn’t wait for instructions. They ran continuously in the background, optimizing based on user preferences. Commerce was no longer a series of discrete human decisions but a continuous optimization process running 24/7 for every connected consumer. By March 2027, the average American was consuming 400,000 tokens per day—a 10x increase from late 2026.
The next link in the chain was already breaking.
Intermediaries.
For the past fifty years, the U.S. economy had built a massive rent-extraction layer on top of human limitations: things take time, patience runs out, brand familiarity substitutes for due diligence, and most people would rather accept a spread than click a few more times. Trillions of dollars in enterprise value depended on these constraints persisting.
It started simply. Agents eliminated friction.
Passive-renewal subscriptions and memberships, unused for months. Introductory prices that quietly doubled after the trial period. Each was redefined as a “hostage situation” that agents could negotiate out of. Customer lifetime value—the metric the entire subscription economy was built on—declined noticeably.
Consumer agents began changing how almost every consumer transaction worked.
Humans don’t have time to compare prices across five platforms to buy a box of protein bars. Machines do.
Travel booking platforms were early casualties because they were the simplest. By Q4 2026, our agents could assemble complete itineraries (flights, hotels, ground transport, points optimization, budget constraints, refunds) faster and cheaper than any platform.
Insurance renewals—the entire renewal model relying on policyholder inertia—were reformed. Agents that re-quoted you annually dismantled the 15-20% premium insurers earned from passive renewals.
Financial advice. Tax preparation. Routine legal work. Any category where a service provider’s value proposition was essentially “I’ll handle the complicated stuff you find annoying” was disrupted, because agents found nothing annoying.
Even areas we thought were protected by the value of human relationships proved fragile. Real estate, where buyers had tolerated 5-6% commissions for decades due to information asymmetry between agents and consumers. But when AI agents equipped with MLS access and decades of transaction data could instantly replicate the entire knowledge base, it all collapsed. A sell-side report from March 2027 was titled “Agent-on-Agent Violence.” The median buyer’s commission in major metro areas had already compressed from 2.5-3% to under 1%, and an increasing number of transactions were completed without a human buyer’s agent.
We overestimated the value of “relationships.” It turned out that what many people called a relationship was just friction with a friendly face.
This was just the beginning of the intermediation layer’s disruption. Successful companies had spent billions to exploit characteristics of consumer behavior and human psychology that no longer mattered.
Machines optimizing for price and fit don’t care about your favorite app or the website you’ve habitually opened for the past four years, nor are they drawn to a carefully designed checkout experience. They don’t get tired and settle for the easiest option, and they don’t default to “I’ve always booked with this one.”
This destroys a special kind of moat: habitual intermediation.
DoorDash (DASH US) is the archetype.
Coding agents have lowered the barrier to launching a food delivery app to nearly zero. A competent developer can deploy a fully functional competitor in a few weeks, and dozens have done so, poaching DoorDash and Uber Eats drivers by passing 90-95% of delivery fees directly to them. Multi-platform dashboards let gig workers track orders across twenty or thirty platforms simultaneously, eliminating the lock-in effects incumbents relied on. Markets fragmented overnight, and margins compressed to near zero.
Agents accelerate destruction in both directions. They empower competitors, then they use them. DoorDash's moat literally was "you're hungry, you're lazy, this is the app on your home screen." Agents don't have home screens. They check DoorDash, Uber Eats, the restaurant's own website, and twenty new vibe-coded alternatives, picking the lowest fee and fastest delivery.
Habitual app loyalty—the foundation of the entire business model—simply doesn't exist for machines.
There's a certain poetry here, perhaps the only instance in this entire story where agents helped the white-collar workers they were about to displace. When they end up as delivery drivers, at least half their income no longer flows to Uber and DoorDash. Of course, this kindness from technology didn't last long—autonomous vehicles became ubiquitous soon after.
The Unraveling of Payment Networks
Once agents controlled transactions, they started looking for bigger paperclips.
Price comparison and aggregation can only do so much. The biggest way to save users money repeatedly (especially as agents start transacting with each other) is to eliminate fees. In machine-to-machine commerce, the 2-3% card interchange fee became an obvious target.
Agents began seeking faster, cheaper alternatives to card networks. Most settled on stablecoins over Solana or Ethereum L2s, with near-instant settlement and transaction costs in fractions of a cent.
MASTERCARD Q1 2027: Net revenue +6% YoY; spend growth slowed to +3.4% from +5.9% last quarter; management cited "agent-led price optimization" and "pressure in discretionary categories" | Bloomberg, April 29, 2027
Mastercard's Q1 2027 earnings became the inflection point. Agentic commerce shifted from a product narrative to an infrastructure narrative. MA fell 9% the next day. Visa also fell but narrowed losses after analysts noted its stronger positioning in stablecoin infrastructure.

Agentic commerce bypassing interchange fees poses a greater risk to card-focused banks and monoline issuers—they capture the majority of the 2-3% fee and have built entire business lines around merchant-subsidized rewards programs.
American Express (AXP US) was hit hardest—white-collar layoffs eroded its customer base, and agents bypassing interchange fees eroded its revenue model, a double blow. Synchrony (SYF US), Capital One (COF US), and Discover (DFS US) all fell more than 10% in subsequent weeks.
Their moat was made of friction. And friction is going to zero.
Macro Transmission
Throughout 2026, the market treated AI's negative effects as an industry story. Software and consulting were crushed, payments and other "toll booths" wobbled, but the broader economy seemed fine. The labor market was softening but not collapsing. The consensus view was that creative destruction is part of any technology innovation cycle. There would be localized pain, but AI's overall net positive effect would outweigh the negatives.
Our January 2027 macro memo argued this was the wrong mental model. The US economy is a white-collar services economy. White-collar workers make up 50% of employment and drive roughly 75% of discretionary consumer spending. The businesses and jobs AI is eating are not the periphery of the US economy—they are the US economy itself.
"Technology destroys jobs, then creates more." That was the most popular and persuasive rebuttal at the time. It was popular and persuasive because it had been true for two hundred years. Even if we couldn't imagine what the future jobs would be, they would surely appear.
ATMs made branch operations cheaper, so banks opened more branches, and teller employment grew for the next two decades. The internet disrupted travel agencies, yellow pages, and physical retail, but invented entirely new industries in their place, creating new jobs.
However, every new job required a human to perform it.
AI is now a general intelligence that keeps improving at the very tasks humans might have transitioned into. A laid-off programmer can't simply pivot to "AI management" because AI can already do that.
Today, AI agents handle research and development tasks that once took weeks. Exponential growth crushes our perception of what's possible, even as Wharton professors try to fit the data into a new S-curve each year.

They write almost all the code. The best of them are vastly smarter than almost all humans at almost everything. And they keep getting cheaper.
AI creates new jobs. Prompt engineers. AI safety researchers. Infrastructure technicians. Humans are still in the loop, coordinating or curating taste at the highest levels. But for every new role AI creates, it eliminates dozens. The new roles pay a fraction of the old ones.
US JOLTS: Job openings fell below 5.5 million; the ratio of unemployed to openings climbed to about 1.7, the highest since August 2020 | Bloomberg, October 2026
Hiring rates had been weak all year, but the October 2026 JOLTS data provided hard evidence. Job openings fell below 5.5 million, down 15% year-over-year.
INDEED: Hiring in software, finance, and consulting down sharply; "productivity enhancement programs" spreading | Indeed Hiring Lab, November-December 2026
White-collar job openings are collapsing, while blue-collar jobs (construction, healthcare, trades) remain relatively stable. What's being replaced are the jobs that write memos (for some reason we're still in business), approve budgets, and lubricate the middle layers of the economy. But real wage growth for both groups was negative for most of the year and declining.
The stock market still paid more attention to GE Vernova's turbine capacity being sold out through 2040 than to JOLTS. The market seesawed between negative macro news and positive AI infrastructure headlines, trading sideways.
The bond market (always smarter than the stock market, or at least less romantic) began pricing in a consumer shock. The 10-year yield drifted from 4.3% down toward 3.2% over four months. But the headline unemployment rate didn't blow out, and its composition details were still ignored by some.
In a normal recession, the cause eventually self-corrects. Overbuilding leads to a slowdown in construction, which leads to lower rates, which leads to new construction. Inventory gluts lead to destocking, then restocking. The cycle mechanism contains the seeds of its own recovery.
The cause of this cycle is not cyclical.

AI gets better and cheaper. Companies lay off workers and use the savings to buy more AI capability, which leads to more layoffs. Laid-off workers spend less. Consumer-facing companies see sales decline, their operations deteriorate, and they increase AI investment to protect margins. AI gets better and cheaper.
A feedback loop with no natural brake.
Intuitively, falling aggregate demand should slow AI buildout. But it doesn't, because this isn't hyperscaler CapEx. It's OpEx substitution. A company that used to spend $100 million a year on people and $5 million on AI now spends $70 million on people and $20 million on AI. AI spending doubles, but within a shrinking total operating cost. Every company's AI budget grows while total spending contracts.
Ironically, the AI infrastructure complex continues to perform even as the economy it's disrupting deteriorates. NVDA still posts record revenue. TSM still runs at 95%+ utilization. Hyperscalers still spend $150-200 billion per quarter on data center CapEx. Economies that purely benefit from this trend, like Taiwan and South Korea, outperform significantly.
India is the opposite. Its IT services sector exports over $200 billion annually, the largest contributor to India's current account surplus and the offset that finances its persistent goods trade deficit. The entire model is built on a value proposition: Indian developers cost a fraction of their US counterparts. But the marginal cost of AI coding agents has dropped to essentially the cost of electricity. TCS, Infosys, and Wipro saw accelerating contract cancellations throughout 2027. The rupee depreciated 18% against the dollar in four months as the services trade surplus that underpinned India's external accounts evaporated. By Q1 2028, the IMF had begun "preliminary discussions" with New Delhi.
The engine of disruption improves every quarter, which means disruption accelerates every quarter. There is no natural floor for the labor market.
In the US, we no longer ask when the AI infrastructure bubble will burst. We ask: what happens to a consumer-credit economy when consumers are being replaced by machines?
2027: The Macro Story Gets Less Subtle
2027 was the year the macro story stopped being subtle. The transmission mechanism of those scattered but clearly negative developments over the previous twelve months became obvious. You didn't need to dig through BLS data; one dinner with friends was enough.
Displaced white-collar workers didn't sit idle. They migrated downward. Many took low-wage service and gig jobs, increasing labor supply in those sectors and further depressing wages there.
We have a friend who was a senior product manager at Salesforce in 2025. Title, health insurance, 401k, $180,000 a year. She lost her job in the third round of layoffs. After six months of searching, she started driving for Uber. Income dropped to $45,000. The point isn't the individual story; it's the math of the second-order effects. Multiply this dynamic by hundreds of thousands of workers across every major metro area. An influx of overqualified labor into services and the gig economy depresses wages for existing workers who were already struggling. Industry-specific disruption diffuses into economy-wide wage compression.

The remaining pool of human-centric jobs faces another correction, happening as we write—autonomous delivery and self-driving vehicles are hitting the gig economy that absorbed the first wave of displaced workers.
By February 2027, it was clear that employed professionals were spending as if they were next. They were working frantically (mostly with AI) just to avoid layoffs, and hopes for promotions and raises had evaporated. Savings rates rose, consumption weakened.
The most dangerous thing was the lag. High earners sustained two or three quarters of apparent normalcy with above-average savings. Hard data only confirmed what was already old news in the real economy. Then came the data point that shattered the illusion.
US initial jobless claims surged to 487,000, the highest since April 2020 | Department of Labor, Q3 2027
Initial claims spiked to 487,000, the highest since April 2020. ADP and Equifax confirmed the vast majority of new claims came from white-collar professionals.
The S&P fell 6% the following week. Negative macro began winning the tug-of-war.
In a normal recession, job losses are broadly distributed. Blue-collar and white-collar workers share the pain roughly in proportion to their employment share. The consumption shock is also broadly distributed and shows up quickly in the data because lower-income workers have a higher marginal propensity to consume.
In this cycle, job losses are concentrated in the top decile of the income distribution. They represent a relatively small share of total employment but drive a disproportionately large share of consumer spending. The top 10% of earners account for over 50% of all US consumer spending. The top 20% account for about 65%. They are the ones buying homes, cars, vacations, dining out, paying private school tuition, and renovating houses. They are the demand base of the entire discretionary consumer economy.
When these workers lose their jobs or accept a 50% pay cut to move to available roles, the consumption shock is enormous relative to the number of unemployed. A 2% decline in white-collar employment translates to roughly a 3-4% shock to discretionary consumer spending. Unlike blue-collar unemployment (where a factory worker is laid off and stops spending the following week), white-collar unemployment has a lagged but deeper impact because they have savings buffers that can sustain consumption for months before behavior changes.
By Q2 2027, the economy had entered a recession. The NBER won't officially date the start for months (they never do), but the data is clear—we've had two consecutive quarters of negative real GDP growth. But this isn't a "financial crisis"... not yet.
The Private Credit Collapse
Private credit grew from under $1 trillion in 2015 to over $2.5 trillion by 2026. A significant share was deployed in software and tech deals, many of them leveraged buyouts of SaaS companies valued on assumptions that revenue would grow at double digits forever.
Those assumptions died somewhere between the first agentic coding demo and the software crash of Q1 2026, but the marks didn't seem to know they were dead.
Public market SaaS companies have fallen to 5-8x EBITDA, while PE-backed software companies still sit on balance sheets at marks reflecting acquisition valuations—based on revenue multiples that no longer exist. Managers are slowly marking down, 100 cents, 92, 85, while the public market comps are saying 50.
MOODY'S downgrades 14 issuers on $18 billion of PE-backed software debt, citing "persistent revenue headwinds from AI-driven competitive disruption"; largest single-sector rating action since energy in 2015 | Moody's Investors Service, April 2027
Everyone remembers what happened after the downgrades. The veterans have seen the 2015 energy downgrade playbook.
Software-backed loans started defaulting in Q3 2027. PE portfolio companies in information services and consulting followed. Several multi-billion dollar LBOs of well-known SaaS companies entered restructuring.
Zendesk was the smoking gun.
ZENDESK breaches debt covenants as AI-driven customer service automation erodes ARR; $5 billion direct lending facility marked to 58 cents; largest private credit software default on record | Financial Times, September 2027
In 2022, Hellman & Friedman and Permira took Zendesk private for $10.2 billion. The debt package was $5 billion in direct lending, led by Blackstone with Apollo, Blue Owl, and HPS participating, the largest ARR-backed loan in history at the time. The loan was explicitly predicated on the assumption that Zendesk's annual recurring revenue would remain recurring. At roughly 25x EBITDA leverage, it only made sense if that assumption held.
By mid-2027, it didn't.
AI agents had been handling customer service autonomously for nearly a year. The category Zendesk defined (ticketing, routing, managing human support interactions) had been replaced by systems that solved problems without generating tickets. The annual recurring revenue the loan was based on was no longer "recurring"—it was just revenue that hadn't left yet.
The largest ARR-backed loan in history became the largest private credit software default on record. Every credit desk asked the same question simultaneously: who else has persistent headwinds disguised as cyclical headwinds?
But the initial consensus had a point: this should have been survivable.
Private credit isn't 2008 banking. The entire architecture is designed to avoid forced selling. These are closed-end funds with locked-up capital. LPs are committed for seven to ten years. No depositors can run, no repo financing can be pulled. Managers can sit on impaired assets, work them gradually, wait for recoveries. Painful, but manageable. The system is designed to bend, not break.
Executives at Blackstone, KKR, and Apollo cited 7-13% exposure to software assets. Manageable. Every sell-side report and FinTwit credit analysis account said the same thing: private credit has permanent capital. They can absorb losses that would blow up levered banks.
Permanent capital. The phrase appeared in every earnings call and investor letter, meant to soothe. It became a mantra. And like most mantras, nobody looked at the fine print. Here's what it actually meant…
Over the past decade, large alternative asset managers acquired life insurance companies, turning them into funding conduits. Apollo bought Athene. Brookfield bought American Equity. KKR took Global Atlantic. The logic was elegant: annuity deposits provided a stable base of long-duration liabilities. Managers deployed those deposits into their own originated private credit, making money twice—spread on the insurance side, management fees on the asset management side. A fee-on-fee perpetual motion machine, working perfectly under one condition.
Private credit had to be safe.
Losses hit balance sheets built to hold illiquid assets against long-duration obligations. That "permanent capital" that was supposed to make the system resilient wasn't some abstract pool of patient institutional money and sophisticated risks taken by sophisticated investors. It was American household savings—"Main Street" money—structured as annuities, allocated to the same PE-backed software and tech assets that were now defaulting. The locked-up capital that couldn't run was life insurance policyholder money, and the rules there are different.
Insurance regulators have been docile—complacent, even—compared to the banking system, but this was a wake-up call. They were already uneasy about private credit concentration at life insurers, and now began to re-risk these assets' capital treatment. This forced insurers to either raise capital or sell assets, neither achievable on attractive terms in a market that was already frozen.
New York, Iowa regulators tighten capital treatment for certain privately rated credit held by life insurers; NAIC guidance expected to increase RBC factors and trigger additional SVO scrutiny | Reuters, November 2027
When Moody's placed Athene's financial strength rating on negative outlook, Apollo's stock fell 22% in two trading days. Brookfield, KKR, and others followed.
Things got more complicated from there. These firms hadn't just created the insurance perpetual motion machine; they'd built sophisticated offshore architectures to maximize returns through regulatory arbitrage. US insurers issued annuities, then ceded the risk to their own Bermuda or Cayman reinsurance affiliates—set up to take advantage of more flexible regulation that allowed holding less capital against the same assets. The affiliate raised external capital through offshore SPVs, creating another layer of counterparties, all investing alongside the insurer in private credit originated by the same parent's asset management arm.

The rating agencies (some of which are themselves PE-held) were not paragons of transparency (unsurprising to nearly no one). The web of interconnections between different firms and different balance sheets was shockingly opaque. When the underlying loans defaulted, the question of who actually bore the loss was genuinely unanswerable in real time.
The November 2027 crash marked the shift from "maybe this is just a normal cyclical downturn" to something more unsettling. "A correlated domino chain betting on white-collar productivity growth"—Fed Chair Kevin Warsh, FOMC emergency meeting, November 2027.
See, it's never the losses themselves that trigger a crisis. It's the recognition of the losses. And there's another, much larger and more important financial domain that we're increasingly afraid of that recognition moment for.
The Mortgage Problem
ZILLOW Home Value Index San Francisco -11% YoY, Seattle -9%, Austin -8%; FANNIE MAE flags "elevated early-stage delinquency rates" in zip codes where tech/finance employment exceeds 40% | Zillow / Fannie Mae, June 2028
This month, the Zillow Home Value Index shows San Francisco down 11% year-over-year, Seattle 9%, Austin 8%. That's not the only worrying headline. Last month, Fannie Mae flagged elevated early-stage delinquency rates in zip codes dense with jumbo loans—borrowers with 780+ credit scores, the kind typically considered "bulletproof."
The US residential mortgage market is roughly $13 trillion. Mortgage underwriting is built on a fundamental assumption: the borrower will remain employed at roughly their current income level for the life of the loan. For most mortgages, that means thirty years.
The white-collar employment crisis threatens this assumption with a persistent downward shift in income expectations. We now have to ask a question that would have sounded absurd three years ago—are prime mortgages safe?
Every mortgage crisis in American history has been driven by one of three things: speculative excess (like 2008 lending to people who couldn't afford homes), interest rate shock (like the early 1980s when rising rates made ARMs unaffordable), or regional economic shocks (like the Texas oil bust of the 1980s or the Michigan auto crisis of 2009).
None of these apply here. The borrowers in question are not subprime. They have FICO scores of 780. 20% down. Clean credit histories, stable employment records, income verified and documented at origination. They are the borrowers every risk model in the financial system treats as the bedrock of credit quality.
In 2008, the loans were bad from day one. In 2028, the loans were good from day one. The world just… changed after origination. People borrowed against a future they can no longer believe in.

In 2027, we flagged early signs of hidden stress: HELOC draws, 401(k) withdrawals, and credit card debt spiking while mortgage payments remained current. As layoffs, hiring freezes, and bonus cuts hit, these prime households saw their debt-to-income ratios double.
They were still making their mortgage payments, but only by stopping all discretionary spending, draining savings, and deferring all home maintenance and improvements. They were technically current on their mortgages, but one shock away from trouble, and the trajectory of AI capabilities suggested that shock was coming. Then we saw defaults start to spike in San Francisco, Seattle, Manhattan, and Austin, even as national averages remained within historical norms.
We are now in the most stressful phase. Home price declines are manageable when marginal buyers are healthy. Here, the marginal buyers face the same income damage.
Concern is building, but we are not yet in a full-blown mortgage crisis. Default rates are elevated but still far below 2008 levels. The real threat is the trajectory.

The intelligent substitution spiral now has two financial accelerators amplifying the real economy downturn.
Labor substitution, mortgage concerns, private market turmoil. Each reinforces the others. Traditional policy tools (rate cuts, QE) can address the financial engine, but not the real economy engine, which is driven not by tight financial conditions. It's driven by AI making human intelligence less scarce, less valuable. You can cut rates to zero, buy every MBS and every defaulted software LBO debt…
But that doesn't change the fact that a Claude agent can do the work of a $180k/year product manager for $200/month.
If these concerns materialize, the mortgage market will crack in the second half of this year. In that scenario, we expect the current equity drawdown to ultimately rival the Global Financial Crisis (57% peak-to-trough). That would put the S&P 500 at roughly 3500—levels not seen since November 2022, a month before the ChatGPT moment.
What is clear is that the income assumptions underlying $13 trillion in residential mortgages are structurally impaired. What is unclear is whether policy can intervene before the mortgage market fully digests what this means. We are hopeful, but cannot deny reasons not to be.
The Policy Dilemma
The first negative feedback loop is in the real economy: AI capability increases → wage expenditure contracts → consumption weakens → profit margins tighten → companies buy more AI → capabilities continue to improve. Then it shifts to finance: income damage impacts mortgages → bank losses tighten credit → wealth effects shatter → the feedback loop accelerates. Both are exacerbated by inadequate government policy responses—the government, frankly, looks confused.

This system was not designed for such a crisis. The federal government's revenue base is essentially a tax on human time. People work, companies pay them, and the government takes a cut. Personal income tax and payroll tax are the backbone of fiscal revenue in normal years.
As of the first quarter of this year, federal revenue is 12% below the CBO baseline forecast. Payroll tax revenue has fallen because fewer people are employed at previous wage levels. Income tax revenue has fallen because people's earnings are structurally lower. Productivity is surging, but the gains flow to capital and compute, not labor.
Labor's share of GDP has fallen from 64% in 1974 to 56% in 2024—a four-decade slow decline driven by globalization, automation, and the persistent erosion of worker bargaining power. In the four years since AI began its exponential progress, that share has dropped to 46%. The steepest decline on record.

Output is still there. But it no longer flows back to businesses through households, which means it no longer flows through the IRS either. The circular flow is breaking, and the government is expected to step in and fix it.
As with every recession, spending increases while revenue declines. What's different this time is that the spending pressure is not cyclical. Automatic stabilizers were designed for temporary unemployment, not structural replacement. The system is paying out benefits that assume workers will be reabsorbed. Many will not—at least not at anything close to their previous wages. During COVID, the government readily accepted a 15% deficit rate, but that was understood as temporary. The people needing government support today were not struck down by a pandemic from which they could recover. They have been replaced by a technology that continues to advance.
The government needs to transfer more money to households while receiving less tax from them.
The United States will not default. It prints the currency it spends, the same currency used to repay lenders. But the strain has already appeared elsewhere. Municipal bonds have shown a troubling divergence so far this year. States without income taxes are fine, but general obligation municipal bonds issued by states that rely on income taxes (mostly blue states) have begun to price in some default risk. Politicians have quickly sniffed an opportunity, and the debate over who should be bailed out is already unfolding along party lines.
To the government's credit, it recognized the structural nature of the crisis relatively early and began discussing a bipartisan proposal they call the "Economic Transformation Act": a framework for direct transfers to displaced workers, funded by deficit spending and a proposed tax on AI inference compute.
The most aggressive proposal on the table goes further. The "Shared AI Prosperity Act" would establish a public claim on the returns from intelligence infrastructure—somewhere between a sovereign wealth fund and an AI output royalty—with the proceeds used to fund household transfers. Private sector lobbyists have already flooded the zone with warnings of a slippery slope.
The politics behind the discussion are frustratingly predictable, amplified by grandstanding and brinkmanship. The right calls transfers and redistribution Marxism, warning that taxing compute would cede the lead to China. The left warns that a tax bill drafted with the help of incumbents would become another form of regulatory capture. Fiscal hawks point to unsustainable deficits. Doves cite the premature austerity imposed after the GFC as a cautionary tale. In the run-up to this year's presidential election, the divisions only deepen.
While politicians squabble, the social fabric is unraveling faster than the legislative process.
The "Occupy Silicon Valley" movement is a microcosm of broader discontent. Last month, protesters blockaded the entrances to Anthropic and OpenAI's San Francisco offices for three weeks. Their numbers are growing, and the demonstrations are drawing more media attention than the unemployment data that sparked them.
It's hard to imagine the public hating anyone more than they hated bankers after the GFC, but AI labs are getting close. And from the public's perspective, with good reason. The founders and early investors of these companies have accumulated wealth at a pace that makes the Gilded Age look pale. The gains from the productivity boom have flowed almost entirely to the owners of compute and the shareholders of the labs running on it, pushing inequality in the United States to unprecedented levels.
Every side has its villain, but the real villain is time.
The pace of AI capability evolution has outstripped the speed of institutional adaptation. Policy responses move at the speed of ideology, not reality. If the government does not reach a consensus on the problem soon, the feedback loops will write the next chapter for them.
The End of the Scarcity Premium
For the entire history of modern economics, human intelligence has been the scarce factor. Capital is abundant (or at least replicable). Natural resources are finite but substitutable. Technological progress was slow enough that humans could adapt. Intelligence—the ability to analyze, decide, create, persuade, and coordinate—was the thing that could not be mass-produced.
The inherent premium on human intelligence derived from its scarcity. Every institution in our economy, from labor markets to mortgage markets to tax codes, was designed for a world where that assumption held.
We are now witnessing the dissolution of that premium. Machine intelligence is now a competent and rapidly improving substitute for human intelligence on an increasing number of tasks. The financial system, optimized for decades for a world of scarce human minds, is repricing. This repricing is painful, disorderly, and far from complete.
But repricing is not collapse.
The economy can find a new equilibrium. Getting there is one of the few remaining tasks that only humans can perform. We need to get it right.
For the first time in history, the economy's most productive asset is generating fewer jobs rather than more. No one's framework applies, because no framework was designed for a world where a scarce factor becomes abundant. So we must create new frameworks. Whether we can build them in time is the only question that matters.
But you are not reading this in June 2028. You are reading it in February 2026.
The S&P is near all-time highs. The negative feedback loops have not yet begun. We are certain that some of these scenarios will not come to pass. We are equally certain that machine intelligence will continue to accelerate. The premium on human intelligence will narrow.
As investors, we still have time to assess how much of our portfolios is built on assumptions that will not survive the end of this decade. As a society, we still have time to act proactively.
The canary is still alive.
Acknowledgments: Thanks to Sam Koppelman of Hunterbrook for copyediting. Our co-author, Alap Shah of LOTUS, contributed the idea for this article—CitriniResearch wrote this section, and he wrote other pieces in the "Intelligence Explosion" series, which are highly recommended and can be found here.