The post-1945 social contract is going to die in public, and the institutions built around it will be the last to notice.
A note before we begin
This is the first piece in a series I am going to write on AI and what it does to the financial, political, and institutional architecture we have inherited. I have been writing in scattered pieces about labor displacement, the middleware trap, private credit’s retreat from software, and the geopolitics of compute, but the more I work through these threads individually the more it becomes clear that they share a single underlying argument that deserves to be made systematically. So I am going to do that, across what I think will be four or five pieces over the coming months, starting here.
The reason to start with the cascade scenario rather than with something more tractable is that I think most of the policy and capital allocation conversation right now is operating under assumptions about pace and reversibility that are not going to hold. People are debating whether AI displaces five percent or fifteen percent of knowledge work, whether the adjustment takes ten years or twenty, whether the affected workers transition into “AI-augmented” roles or get pushed into a different sector entirely. These are interesting questions, but they are not the most important ones. The most important question is what happens to the financial and political system if the adjustment is faster, broader, and more synchronized than any of the comfortable scenarios assume, and whether the institutions we have are capable of managing that adjustment competently.
So this piece is the stress test. The pieces that follow will be more granular: the specific mechanics of credit market deterioration, the political economy of AI taxation, the geography of the recovery, the implications for private credit strategy specifically, and what the new social contract might actually look like in operational terms. But you have to see the whole shape of the thing before the parts make sense, and the whole shape is what I am going to lay out here.
One assumption I am going to make and not defend at length: that the underlying capability is real, and that systems able to perform the substance of a large share of knowledge work either exist now or arrive on something close to the current trajectory. The how and why of that (what changed in the models, why this wave is different from the automation scares that preceded it, and what the productivity statistics do and do not show) deserves more than a paragraph, so it is out of scope here; Part II of this series takes it on directly. Two things are worth saying now. First, the argument does not require artificial general intelligence. It requires that AI absorb enough of the median task content of white-collar jobs that the same output clears at materially lower headcount, which is a much lower bar, and the one the labor data below suggests is already being met. Second, the assumption is doing real work in everything that follows, which means it is also a failure point: if capability stalls, the cascade does not happen. That is part of what makes this a stress test rather than a prophecy.
On the question of “when”
I should address this directly because it is the first question any honest reader will ask. When does this happen?
I do not know, and I want to be straightforward about why.
In January 2026, Sequoia Capital’s Pat Grady and Sonya Huang published a piece titled “2026: This is AGI,” arguing that long-horizon agents capable of sustained autonomous work have crossed the functional threshold that matters, and that we are already inside the AGI era whether we have noticed or not. Their colleague David Cahn, writing one month earlier in December 2025, made the opposite case: that 2026 would be a “year of delays,” that the AGI timeline was being pushed out into the 2030s, and that the supply chain constraints on data center buildout were going to slow the diffusion of capability well below the trajectory the enthusiasts were forecasting. These are two partners at the same firm, looking at the same data, reaching opposite conclusions about whether the transformation is happening now or is still several years away.
If Sequoia cannot agree internally, I am not going to pretend I have superior visibility.
What I do have is direct exposure. I spend part of my week building a B2B2C software product (Vintrak) with these tools, and watching what long-horizon agents do to a development timeline is a large part of why I take the capability curve seriously. The displacement mechanism this series describes is not something I read about. It is something I operate.
What I can say is the following. The first-order labor market signal is already present in the data. One in four American workers who lost their jobs in 2024 worked in professional and business services, per S&P Global’s analysis of BLS turnover data, in a sector that accounts for roughly 13 percent of employment. And the Stanford Digital Economy Lab’s “Canaries in the Coal Mine” research documents a 16 percent relative employment decline since late 2022 for workers aged 22 to 25 in the most AI-exposed occupations, controlling for firm-level shocks, while employment for experienced workers in the same occupations remained stable or grew; in a February 2026 follow-up the authors checked whether interest rates could explain the pattern instead, and found they could not. The phrase being used now is “companies are keeping the pilots and automating the co-pilots.” That is consistent with the early stage of the cascade I describe in this piece. It is also consistent with a much slower transition that never reaches the second-order effects. The data does not distinguish between those scenarios yet, and anyone who tells you otherwise is selling something.
What I think is most likely is that the timing is set by the political calendar as much as by the technological one. The 2026 midterms, now five months out, are going to be contested in conditions that include visible white-collar displacement but not yet the credit deterioration. The New York Fed’s most recent household debt data shows a K-shaped credit market: delinquencies are rising, but the distress is concentrated in subprime and near-prime borrowers, while prime borrowers have barely moved. That is exactly what the cascade model predicts for its first phase, in which severance and savings buffers keep prime credit looking pristine for twelve to fifteen months after the income shock begins. It is also exactly what you would see if there were no cascade at all. The diagnostic that separates the two scenarios is the 30-day delinquency rate on prime credit card receivables, and I will be watching it quarterly. If it ticks up materially through 2027 while headline unemployment stays low, the cascade thesis is being confirmed. If it stays flat, the thesis is wrong, and I will say so. Whichever party loses the midterms will spend the next twenty-four months arguing that the other party mishandled the early stages of the cascade, and the 2028 presidential election will be fought explicitly on the question of who absorbs the loss and how. I think that election is going to be the most consequential in American history since 1932, and possibly since 1860, because the policy choices made in its aftermath will determine which of the post-cascade scenarios I describe in this piece actually obtains. Whether the cascade itself peaks before or after that election is the question I cannot answer.
The honest version of “when” is: the early stages are already visible if you know where to look, the middle stages are dependent on credit cycle dynamics and political response timing that nobody can predict precisely, and the resolution is at least five years out and probably closer to ten. If you are positioning capital, the planning horizon is now, not later. If you are positioning institutionally, you have less time than you think.
With that out of the way, let me walk through what the cascade actually looks like.
The cohort and what makes it different
There is a particular kind of failure mode that financial systems are not built to handle, and we are walking into it with our eyes open. It is not a credit shock. It is not a liquidity shock. It is not even an asset-price shock in the conventional sense. It is a collateral integrity shock, which is what you get when the underlying signal that lenders use to price risk stops meaning what it used to mean. The signal in this case is employment, and the cohort losing it is the one the entire post-1945 American financial architecture was built around: the credentialed, prime-credit, white-collar household.
Most of the AI-and-jobs commentary I read frames this as a labor market story. It is not. Labor markets adjust, painfully but eventually. What is coming is something different. It is the simultaneous deterioration of asset classes that were specifically designed never to deteriorate together, anchored by a borrower base that was specifically considered the safe one.
The BLS classifies roughly 70 million Americans as working in “management, professional, and related occupations,” which is about 44 percent of the workforce. The subset that matters for the cascade scenario is narrower: the urban and inner-suburban professionals concentrated in finance, law, consulting, technology, marketing, corporate operations, and middle management. Call it 40 to 50 million workers with median household incomes between $120,000 and $180,000. These households carry $300,000 to $600,000 mortgages, meaningful auto loan balances, and credit card debt that they service comfortably because of the income, not despite it.
Three structural facts about this cohort matter for what follows.
First, they are the borrower base. The top 10 percent of earners paid roughly 70 percent of all federal income taxes in 2023, and the top 50 percent paid 97 percent. The same concentration applies to prime credit. The mortgage-backed securities market, the prime auto ABS market, the prime credit card receivables market, and the jumbo mortgage market are all built on the assumption that this cohort’s employment is durable. Their FICO 780+ status is not just a statistical artifact. It is the load-bearing wall of half the consumer credit complex.
Second, they are geographically concentrated. The bulk of them live in Boston, the San Francisco Bay Area, the New York metro, Seattle, DC, Austin, Raleigh, and a handful of other knowledge-economy hubs. Real estate prices in these zip codes are not just elevated. They are priced for permanence. The model the housing market implicitly runs on is that knowledge-worker wage growth continues indefinitely and that the supply of credentialed buyers replenishes itself through the educational pipeline.
Third, they are the people who operate the institutions that would manage a crisis. Mid-level Treasury staff. BLS economists. Fed researchers. Congressional staff. Regulatory attorneys. Compliance officers at systemically important banks. This is not a sentimental observation. It is a structural one, and it matters at the end of the story.
The first-order shock
Take half of that cohort offline over twenty-four to thirty-six months. Not all at once, but in a way that compounds: hiring freezes hardening into layoffs, layoffs widening from junior roles into mid-level, mid-level layoffs spreading from technology into financial services into law into corporate operations. The pattern is already legible in the data, as I noted above. What is not yet legible is whether the trend bends back toward equilibrium or continues compounding into the cascade scenario. Both are possible. The argument of this piece is that the second possibility is being underweighted by capital allocators and policymakers.
What happens when 20 to 25 million prime borrowers lose income simultaneously?
The first thing that happens is that nothing visible happens. Severance packages last six to nine months. Households have savings buffers, partner income, home equity lines of credit they can draw on. The unemployment numbers tick up but not catastrophically, and the financial press writes pieces about “sectoral rebalancing.” This is the phase that lasts approximately twelve to fifteen months, during which a lot of capital allocators convince themselves that the adjustment is gradual and orderly. We may be in this phase right now. We may not. The diagnostic is not available until after the fact.
Then severance runs out. The savings buffers thin. The HELOCs get drawn down to limits and frozen. Households start carrying credit card balances they did not previously carry. The 30-day delinquency tick on prime credit card receivables shows up first because the math is fastest there: a household carrying $20,000 in revolving balances at 24 percent APR while underemployed cannot last more than a few months. Then 60-day delinquencies. Then 90-day. By month eighteen, prime auto ABS starts deteriorating. By month twenty-four, jumbo mortgage delinquencies become statistically visible. By month thirty, MBS pools start taking real losses.
The thing that distinguishes this from 2008 is not the magnitude. It is the signal collapse.
Why the system cannot absorb it
2008 was a subprime shock that contaminated prime. The narrative the financial system told itself afterward, and built its regulatory response around, was that prime credit is safe and subprime credit is dangerous and the boundary between them is identifiable and durable. Dodd-Frank, the post-crisis Basel revisions, the stress testing regime, the GSE conservatorship, the QM rule, all of these are built on the assumption that you can sort borrowers into risk buckets using observable signals (FICO, DTI, LTV, employment history) and that those signals carry stable predictive content over time.
This is the assumption that breaks.
When FICO 780+ borrowers with prime employment histories and conforming DTIs start defaulting en masse, the models do not just mispredict. The entire underwriting paradigm becomes retrospectively meaningless. A FICO score is a backward-looking measurement of behavior under a particular set of economic conditions; it is not a forward-looking measurement of solvency under conditions that have never previously existed. Banks cannot price new credit because they cannot identify who is actually creditworthy in a world where five years of stable employment as a senior analyst no longer implies anything about the next eighteen months. Credit markets do not seize because of panic. They seize because of epistemological collapse. Nobody knows what anything is worth, including the lenders, and especially the regulators.
The regional banking system gets hit first and hardest. Regional banks have concentrated commercial real estate exposure in the exact metros that produced the highest-credit borrowers. Their loan books are heavy in jumbo mortgages, professional services CRE, and prime consumer lending. They do not have the diversification or the implicit federal backstop that the systemically important banks do. You get a wave of regional bank failures in the second and third years, mostly resolved through forced sales and FDIC-assisted transactions, but the resolution mechanism strips capital from the surviving institutions and accelerates concentration in the four or five largest banks.
The GSEs are a separate problem. Fannie and Freddie guarantee a huge share of the conforming mortgage market, and their guarantee fees were set assuming a particular default distribution. When the actual distribution shifts violently rightward, the implicit Treasury backstop becomes explicit. The political fight over what that backstop looks like (forced recapitalization? full nationalization? receivership and wind-down?) is going to be one of the defining policy battles of the period.
Pension funds and insurance companies are the slow leak. Their fixed-income allocations are heavy in agency MBS, prime auto ABS, and investment-grade corporates. They do not mark to market the way banks do, so the damage shows up gradually, but it shows up. State pension funds with already-marginal funding ratios start missing their assumed rates of return by hundreds of basis points. The municipal solvency question, which was already an open one in places like Illinois and New Jersey, becomes acute.
The housing leg
The geographic concentration of the cohort is what turns the housing piece from a recession into a structural repricing. Boston, the Bay Area, NYC metro, Seattle, DC, Austin, Raleigh. These markets did not price in trend wage growth. They priced in exceptional wage growth for a specific category of worker, indefinitely. The Cambridge condo at $1.6 million and the Palo Alto bungalow at $3.2 million and the Brooklyn Heights brownstone at $4 million were not priced by people who thought tech and finance compensation would continue at trend. They were priced by people who thought it would continue accelerating.
When the buyer pool contracts because the buyers no longer have the income to qualify and the existing owners start needing to sell because their own income disappeared, the price discovery process is brutal. I would put the realistic range at 35 to 50 percent nominal declines in the most affected zip codes within eighteen months of the cascade becoming visible. Not nationally. Locally. The Bay Area and Boston take the worst of it because their economies are most concentrated in the affected sectors. Austin and Raleigh follow because the migration thesis that pumped them up reverses. NYC and DC bottom out somewhere in the middle because their economies are more diversified, but their condo markets get gutted.
The second-order effect is strategic default. A household carrying a $1.2 million mortgage on a property now worth $700,000, and watching their neighbors walk away, starts asking why they are continuing to service a loan that no longer makes economic sense. Strategic default is rational, and rationality is contagious. The wave of it among underwater-but-still-employed borrowers is what extends the damage to MBS pools that would otherwise have survived the layoff wave.
The third-order effect is municipal. Boston, San Francisco, New York, and DC fund themselves heavily through property taxes and the income taxes of high earners. When property values collapse and high earners’ incomes collapse simultaneously, the cities that produced the highest-credit borrowers in the country become the cities that cannot fund basic services. Transit cuts, school cuts, public safety cuts, deferred maintenance becoming permanent. The exodus of those who can move accelerates the spiral. By year three, you have a fiscal crisis in the exact metros that were considered economically invulnerable two years earlier.
The second-order job losses
The replaced workers were the demand base for an enormous service economy. Restaurants. Personal services. Premium retail. Private schools. Home renovation. Healthcare elective spend. Domestic travel. Boutique fitness. Personal training. Dog walking. The whole ecosystem of urban professional consumption.
That is another 15 to 25 million jobs that do not get automated but disappear because their customers did.
These workers generally have worse credit, fewer savings, and a faster path to insolvency than the cohort that lost knowledge work. They also have substantially less political voice. What you get is a working-class crisis stacked on top of a professional-class crisis, and at first the policy interests are not aligned: the professional class wants debt relief, mortgage forbearance, and AI taxation, while the service class wants direct income support and protection from immigration competition that has nothing to do with the actual cause of their job loss. In the early phase, both groups blame each other before they blame the firms that automated the upstream demand source, because that is what American politics has trained them to do.
But the cascade carries a feature that 2008 did not, and it may end up mattering more than the interest divergence. In 2008, blame was contested and diffuse; bankers, borrowers, regulators, and rating agencies all took a share, and the resentment ran horizontally between classes as much as upward. Here the cause is unusually legible and the beneficiaries are unusually few. The laid-off consultant and the laid-off waitress can point at the same thing, and the thing has a name, a stock price, and a small number of identifiable owners. Shared misery with a common villain is the raw material of cross-class coalition, something American politics has not produced at scale since the 1930s. The early fragmentation and the later convergence are both real, and the sequence from one to the other is the dynamic that produces the political phase change I will get to in a moment.
The fiscal mechanics
Run the numbers. The top 10 percent of earners paid roughly 70 percent of federal income taxes; the top 5 percent paid 60 percent. If you halve the income of even a quarter of that cohort, federal individual income tax revenue drops on the order of 20 to 30 percent in a single fiscal year. For scale, individual income tax receipts fell roughly 20 percent in fiscal 2009, and that was a shock centered on capital gains and subprime-adjacent employment, not on the cohort that pays most of the tax. Simultaneously, the automatic stabilizers (UI, SNAP, Medicaid, ACA subsidies) spike. In recent recessions these added roughly 1 to 2 percent of GDP to the deficit, but those were recessions where the affected workers were not the primary source of tax revenue. In this scenario, the revenue side and the spending side both move violently in the wrong direction at once.
Realistic deficit numbers are 15 to 20 percent of GDP. For comparison, the pandemic-year deficit peaked around 15 percent. The difference is that the pandemic deficit was a transitory shock that markets and the Fed could see through. This one is not transitory, and pricing it as such requires a view about what the post-cascade economy looks like that nobody has yet developed.
Treasury auctions still clear. There is nowhere else for global capital to go on this scale, and the dollar’s status as the reserve currency continues to function precisely because the alternatives are worse. But the fiscal space to actually respond to the crisis evaporates. There is no room for the kind of stimulus that papered over 2008 and the pandemic. The Fed can buy assets, but it cannot replace the income of 25 million white-collar households. The political fight becomes about who absorbs the loss, not about how to make the loss go away.
State and local fiscal mechanics are worse. No central bank backstop. No money printing. Immediate service cuts. The states with the heaviest knowledge-economy exposure (California, New York, Massachusetts, Washington) take the worst of it because their income tax structures are progressive and their high earners pay the bulk of the revenue. The states with the lowest exposure (the Sun Belt, the Mountain West, the resource-extraction economies) come through structurally better positioned, which has political consequences I will return to.
The view from Greater Boston
I live here, so let me make this concrete.
If you set out to design a metropolitan economy to take the full force of the cascade, you would design Greater Boston. Professional services, finance, technology, and venture-funded biotech as the export base. Higher education as both the dominant cultural institution and a top employer. A housing market priced for the indefinite continuation of knowledge-economy wage growth. And underneath all of it, a municipal finance structure more dependent on property taxes than any major city in the country, in a state that forbids its cities from taxing anything else.
Boston’s fiscal stress test has already started, and AI had nothing to do with it. Property taxes fund roughly 71 percent of the city’s $4.8 billion budget, and commercial property alone supplies more than a third of city revenue, against about 11 percent in a typical American city. The remote-work shock alone pushed office assessed values down 9 percent in fiscal 2025, and the Boston Policy Institute and Tufts’ Center for State Policy Analysis project declines of 35 to 45 percent from 2024 levels by 2029, a cumulative revenue gap of $1.7 billion or more. Massachusetts precludes local sales and income taxes, so the only release valves are residential rate increases or service cuts. Read that again: the budget math is already broken, and the cascade has not started. Whatever Boston’s commercial tax base looks like after remote work, the cascade scenario subtracts the white-collar incomes and the residential values on top of it.
The lab market is the preview, and the inner suburbs are living in it now. Greater Boston is the largest life science market in the country, and just over a third of its lab space sat vacant at the end of last year. In Somerville and Malden, vacancy is 62 percent. Somerville bet a meaningful share of its commercial tax diversification on lab buildings in Union Square and Boynton Yards that are now competing for tenants in that submarket. Those buildings were not underwritten by people who thought biotech demand would continue at trend. They were underwritten by people who thought it would accelerate indefinitely. That is the same pricing-for-permanence error the housing section of this piece describes, already realized, in commercial form, five years early.
The eds-and-meds anchor that Boston tells itself about is half real. The meds half holds; healthcare delivery is physical, regulated, and demographically guaranteed. The eds half is squarely in the crosshairs of the credentialing collapse I described above, and Greater Boston is the most university-dependent regional economy in America. The universities are not just employers. They are demand engines (students renting apartments, international tuition, parental spending, research overhead) and, more structurally, they are the replenishment pipeline for the credentialed buyer pool that holds up housing values in Cambridge, Somerville, Brookline, Newton, and Arlington. If the bachelor’s degree premium contracts the way I expect, Boston loses the input side and the output side of the same machine.
Two structural wrinkles make the municipal math worse here than elsewhere. First, Proposition 2½ means that when commercial values fall, the levy shifts mechanically onto residential taxpayers; homeowners pick up the tab at precisely the moment their incomes are the thing under pressure. Second, local aid flows from a state income tax paid disproportionately by high earners, topped by a millionaires surtax that is the single most cascade-exposed revenue instrument in the Commonwealth. The local levy and the state backstop break simultaneously. That is the municipal version of the correlated-asset problem this entire piece is about: revenue streams that were assumed to be independent turn out to share a single point of failure, and the point of failure is the white-collar wage.
What would a competent municipal response look like? Four things, none of them expensive yet. Stress test the budget against the combined scenario (commercial and residential assessment declines plus state aid cuts) rather than one variable at a time; the BPI report models the office problem, and nobody has published the version where the residential side moves too. Stop funding recurring spending with new-growth revenue, which is the municipal equivalent of underwriting to peak ARR. Make it structurally cheaper to live here, because a metro that loses high incomes survives on whether middle incomes can afford to stay, and the region’s housing production machinery is currently designed to prevent that. And coordinate regionally, because 101 municipalities each defending their own commercial tax base is exactly the process that produced the lab glut.
The local diagnostics, for anyone in these city halls who wants to watch the same data I am watching: commercial abatement and appeal volume, the new-growth line in the annual levy calculation, April state income tax and surtax receipts, jumbo mortgage delinquencies in Suffolk and Middlesex counties, and lab absorption. None of these requires a new dashboard. All of them are sitting in reports these governments already produce.
Nobody running a Greater Boston city is responsible for the cascade. Every one of them is responsible for whether their budget assumes it cannot happen.
The political phase change
This is where the story stops being an economics problem and becomes something else.
The cohort being displaced is highly educated, articulate, well-networked, and was until very recently the median voter in suburban swing districts in Pennsylvania, Michigan, Wisconsin, Georgia, Arizona, North Carolina, Virginia, Colorado, and Minnesota. They are not going to quietly retrain as HVAC technicians. They are going to organize, because they have always been the people who organize, and they are going to do it with social capital, professional networks, and existing institutional access that the previous wave of displaced workers (manufacturing in the 1980s, retail in the 2010s) did not have.
The political timeline I sketched in the preface matters here. The 2026 midterms are being contested under conditions where the early-stage white-collar displacement is visible but the cascade has not broken through into the credit markets or the housing repricing. The party out of power will run on it. The party in power will downplay it. Whoever wins will be tactically positioned but structurally trapped, because the underlying dynamic is going to keep moving regardless of which party holds Congress. The 2028 presidential election is the one that matters, because by then the cascade is likely to be in its middle innings, the cohort that operates the institutions will be visibly under stress, and the question of who absorbs the loss will be unavoidable.
You get one of two political configurations out of 2028, probably both sequentially.
The first is aggressive redistribution: a panicked Congress passing AI taxation, windfall taxes on firms that automated, forced equity stakes in large AI deployers, capital controls, and some version of universal basic income or a federal job guarantee. The proposals will be popular. They will also be badly structured, because the political timeline does not allow for careful design, and the immediate effect will be capital flight and a sharp contraction in private investment that makes the underlying economic problem worse before it makes it better. This is the conventional first response.
The second is populist realignment, which is what happens after the first response fails to deliver visible relief fast enough. The realignment of 2016 to 2024 is going to look procedural by comparison. You get a politics built around explicit producerism, anti-finance, anti-credentialism, anti-technocracy, with strong cross-class appeal because both the displaced professional class and the working class blame the same identifiable set of corporate actors. The exact ideological flavor is unpredictable. The structural feature (a politics that treats AI deployment as a question of legitimate corporate behavior rather than a market outcome) is not.
What both configurations have in common is that the institutional response capacity is already degraded by the time it is needed. The mid-level Treasury staff, the BLS economists, the Fed researchers, the congressional staff who would normally model the crisis and design the response have been in the displaced cohort. The state capacity to manage a once-in-a-century economic adjustment is itself one of the things being automated. This is the failure mode that turns a manageable transition into something genuinely worse.
What the AI-deploying firms actually experience
The least intuitive part of the cascade is what happens to the firms that drove it.
The first four to eight quarters look spectacular. Margin expansion across the board. SG&A collapses. Revenue per employee figures that look like artillery results. The valuation multiples on the firms that automate fastest go vertical because the productivity story is legible and the costs have already been taken out.
Then the demand wall arrives.
The customers of these firms, directly and indirectly, were the people they replaced. Enterprise SaaS sells to enterprises whose budgets get gutted when their customers’ budgets get gutted. B2B sales cycles extend toward infinity. The ARR per customer figures that defined the previous decade compress as renewals come in at discounts and seat counts shrink. The advertising businesses that depend on consumer spending watch CPMs collapse as ad spend rationalizes. The cloud infrastructure businesses watch usage growth slow as the AI workload boom hits the limits of what enterprises can monetize.
The valuation case for “AI productivity gains” assumes the productivity gets monetized. If aggregate demand contracts faster than costs, you get a deflationary spiral inside the sector that caused it. The firms with pricing power and government customers (defense, healthcare, infrastructure) hold up. The firms with consumer or mid-market business customers do not. The “Magnificent Seven” thesis becomes a “Magnificent Two or Three” thesis very quickly, and the dispersion within the AI complex becomes the dominant alpha source for the next half-decade.
I want to be careful here because this is a contested point. The bullish case is that AI deployment expands the productive frontier so dramatically that the gains overwhelm the demand contraction; the new equilibrium has higher GDP per capita even with higher unemployment because the surplus gets redistributed through fiscal mechanisms. The bearish case is what I have just described. I think the bearish case is more likely on a five-year horizon and the bullish case is more likely on a fifteen-year horizon, but the transition path matters enormously for capital allocation, and the transition path is what this essay is about.
What the world looks like on the other side
Now the part that matters for actually positioning capital. The cascade is not a permanent state. It is a four-to-six-year depression with structural unemployment in the 15 to 20 percent range, and the institutional landscape that emerges on the other side is recognizably different from the one we have now. Here is what I think it looks like.
The social contract gets rewritten, badly at first and then more durably
Some version of universal income or job guarantee passes. The first version is poorly designed, expensive, and capital-flight-inducing. The second version, three to five years in, is better structured: financed through AI deployment taxation, indexed to productivity gains, administered through existing transfer infrastructure (Social Security, the EITC) rather than new bureaucracy. It is not generous by European standards. It is enough to prevent the worst social outcomes and not enough to substitute for productive work. It is paid for by a combination of explicit AI taxation, sharply higher capital gains rates, a financial transaction tax that finally passes after thirty years of failed attempts, and the inflation that comes from monetizing the transition.
The professional middle class does not return to its previous form. The “credential-degree-job-credit-asset” sequence that defined American adulthood from 1945 to 2020 is broken. What replaces it is more bifurcated: a smaller cohort of high-skill specialists who command very high incomes in roles that genuinely cannot be automated (judgment-heavy, relationship-heavy, regulated, or physically embodied), and a much larger cohort of variable-income workers piecing together project work, owner-operated micro-businesses, and transfer income.
The credentialing economy collapses, then partially reforms
The university system is not coming back in its current configuration. The economic premium on a bachelor’s degree that drove four decades of tuition inflation depended on the existence of the knowledge-economy jobs the degree credentialed for; when those jobs contract by half, the demand for the credential collapses with them. The top fifty institutions survive on endowments and graduate professional programs. The next two hundred consolidate, merge, or close. What replaces the system is messier and probably better, and given what higher education means to the regional economy I just described, it deserves its own essay in this series rather than a paragraph here.
The financial system is partially nationalized through forced recapitalization
The regional banking sector consolidates into something more like the European model: four or five large universal banks, a much smaller community banking footprint, and significant federal equity stakes in the survivors. The GSEs are either fully nationalized or wound down in favor of a direct federal mortgage insurance program. The asset management industry consolidates further around a handful of giant passive players and a smaller group of specialists. The middle of the industry (the active mutual funds, the smaller asset managers, the regional wealth management firms) gets squeezed out.
Private credit comes through this paradoxically well. Not the SaaS-direct-lending model, which gets repriced violently along with everything else, but the broader category of non-bank lending to operating businesses with real cash flows. When the banking system is partially seized, the firms that can underwrite credit outside of it command enormous economic rent. The funds that built that capability during the 2010s end up structurally well-positioned. This is not a tactical observation. It is the basis of an entire investment thesis for the recovery period, and one I will develop in much more detail in a later piece in this series.
The geography of the country rearranges
The knowledge-economy metros take the worst of the damage and recover slowly. The Sun Belt and the Mountain West come through structurally better because their economies were less concentrated in the affected sectors and their housing was less mispriced. Texas, Florida, Arizona, Tennessee, North Carolina, Georgia, and Utah end the cascade with larger shares of national GDP, population, and political power than they started with; California, New York, Massachusetts, and Illinois end it materially smaller. The Senate map consequences alone are large and durable, and the full argument (including where the exceptions are, and what it means for capital that can move) is the subject of a dedicated piece later in this series on the new geography of American capital.
A new economic settlement around AI deployment
By year five or six post-cascade, the legal and regulatory framework around AI deployment looks substantially different. Firms above a certain size are required to disclose automation-related employment changes. Large AI deployments require something like an environmental impact statement for labor. The tax code distinguishes between revenue from AI-deployment efficiency gains and revenue from genuinely new economic activity, with the former taxed at meaningfully higher rates. Some form of compulsory profit-sharing or worker equity provision attaches to large-scale automation events. None of this is particularly elegant, and most of it is gameable, but the political demand for some response is overwhelming, and you get the response that the institutions are capable of producing.
The firms that adapt to this framework early end up structurally advantaged. The firms that try to fight it through litigation and regulatory arbitrage end up in much worse positions five years later than they would have been if they had cooperated with the inevitable settlement.
The professional class reconstitutes itself in a different form
This is the part I am least confident about, but the most interested in. The professional cohort that loses its employment base does not disappear. It is highly educated, articulate, networked, and capable of organizing itself economically. What I think emerges, slowly and unevenly, is a new category of professional work that the previous economy did not have categories for: judgment-and-relationship intermediaries who sit between AI systems and the humans who need to use them, regulated certification specialists in domains where AI cannot operate without human signoff (medicine, law, finance, engineering), and a substantial expansion of the small-business and owner-operator economy as professionals who were previously employees become principals of their own micro-firms.
The economic returns to this new category of work are lower on average than the returns to the previous knowledge-economy jobs, but the variance is higher and the top of the distribution is richer. The professional class becomes less salaried and more equity-and-cash-flow-based, more entrepreneurial, more local, and politically more aligned with small-business interests than with large-firm interests. This is a major realignment, and it has consequences for the structure of both major political parties that I will return to in a later piece.
The non-obvious risk
I said earlier that the thing that actually keeps people up at night is not the unemployment number. It is that the replaced cohort includes most of the people who operate the institutions that would manage the crisis. I want to come back to this because it is the thing I am least sure the system has internalized.
When you automate the people whose job is to model and respond to economic shocks, your response capacity degrades exactly when you need it most. The Fed cannot run a complex emergency facility with half the staff it had five years ago, even if AI can in principle do most of the analytical work, because the institutional knowledge required to deploy that work effectively lives in the people, not the tools. The Treasury cannot run a comprehensive bank recapitalization program without the mid-level staff who know how to negotiate with bank holding companies and how to structure the legal documents. Congress cannot pass coherent legislation when the staffers who would draft it are themselves in the displaced cohort and the firms that would lobby on it have automated their government affairs functions.
This is the failure mode that I think is genuinely under-modeled. It is not that the institutions fail. It is that they respond slowly and badly in a situation that demands speed and competence, and the gap between the response that is needed and the response that is delivered is what determines whether the cascade becomes a four-year depression or a seven-year one.
Where the asymmetric positioning is
I am going to be deliberately brief on the investment implications here, because most of what matters is general principle rather than specific recommendation, and I will develop the specifics in subsequent pieces in this series.
The asymmetric positioning is in things that survive both the deflation and the political response. Hard currency exposure outside the affected sectors. Equity in firms that sell to governments rather than to consumers (defense, infrastructure, healthcare delivery, regulated utilities). Operational businesses with real cash flows that are not legible as “AI rents” and therefore do not attract windfall taxation. Physical assets in jurisdictions outside the affected metros. Credit instruments with floating rates and senior security in operating businesses that survive the demand contraction.
The stranded-SaaS thesis I have written about elsewhere actually gets more attractive in this environment, not less, and it is worth spelling out why, because the surface reading runs the other way. The SaaSpocalypse narrative has repriced an entire category indiscriminately. Venture-backed B2B software companies in the $2 to $5 million revenue range cannot raise their next round at any reasonable price, not because each one was examined and found wanting, but because the category-level story turned and growth capital left all at once. Indiscriminate repricing is where mispricing lives. The cascade is selective; the funding withdrawal was not. A meaningful fraction of these companies own real workflows, real customer relationships, and real proprietary data in markets too small or too unglamorous for the AI labs to attack directly, and those assets are exactly how AI capability actually reaches business processes: through the software that already sits inside them.
There is a second reason, and it is the one almost nobody is pricing. The same force that destroyed these companies’ fundraising narrative is collapsing their own cost structure. A company that needed one more round to reach breakeven can increasingly reach it without one, because AI compresses its engineering, support, and sales costs just as surely as it compresses its valuation. The technology that stranded them is also what makes a meaningful subset of them self-sustaining. Capital that can bridge the gap between those two facts, with structure that protects the downside while the sorting happens, is providing liquidity at the moment it is scarcest, which is the oldest asymmetric trade there is. The best venture and credit vintages on record were deployed into dislocations, not recoveries. RavenRock is positioned for this specifically, and I will write about the strategic implications in detail later in this series. Yes, I am talking my book here. The book exists because of the thesis, not the other way around.
And to the reader allocating capital who has gotten this far and is asking why anyone should fund anything if the cascade is coming: notice that this thesis does not require the cascade. The funding withdrawal already happened; the mispricing already exists. In the mild scenario, the survivors get refinanced into a recovering market. In the cascade scenario, they get acquired by the consolidators or become the consolidators. The positioning works in both states of the world, which is the only kind of positioning this piece recommends.
What you want to avoid is the entire complex of bets that depend on the continuation of the previous social and economic settlement: long-dated US consumer credit, residential real estate in the affected metros, large-cap tech equity at current multiples, traditional active asset management franchises, anything that requires the professional middle class to keep spending at trend.
Closing
The cascade scenario is not a forecast. It is a stress test. The actual path is probably less abrupt and less synchronous than I have described, and the system has surprised forecasters with its absorptive capacity many times before. What I am confident about is the direction: the post-1945 American social contract is no longer adequate to the economic forces being unleashed, the institutions built around it are going to be tested in ways they have not previously been tested, and the political and financial architecture on the other side is going to be meaningfully different.
The interesting question is not whether the transition happens. It is whether the transition is managed competently or incompetently, and whether the institutions that emerge on the other side are better or worse than the ones we have now. I am cautiously optimistic that what emerges is better, in the same way that the post-1945 settlement was better than the one it replaced. But the path from here to there is the part of the story that matters, and underestimating how rough that path is going to be is the single biggest mistake I think most capital allocators are currently making.
The pieces that follow in this series will work through the specific mechanics: the credit market dynamics in detail, the political economy of AI taxation, the new geography of American capital, the implications for private credit strategy, and what a functioning post-cascade social contract might actually look like in operational terms. The cascade is the framing. The pieces are the working out.
Part II takes on the most common rebuttal to everything you just read: that the productivity statistics show nothing happening, so nothing is happening. Robert Solow famously observed in 1987 that you could see the computer age everywhere but in the productivity numbers, and his paradox took fifteen years to resolve. It is resolving again right now, faster, and the resolution is not the reassuring part of the story. It is the confirmation. If you want that argument when it publishes, subscribe; it costs nothing, and you will know whether I am right on the same schedule I do.
Plan accordingly.
Jason Mackey is co-founder and Managing Partner of RavenRock, a Boston-based structured bridge capital fund, and is currently building Vintrak, a B2B/2C SaaS product. He writes here about AI-driven economic disruption, politics, and entrepreneurship.
