The monsoon reached Bangkok the same week the delegations did. From a rented desk above Silom Road, I watched the motorcades cut through water that had risen past the curb, and I thought about the peculiar theater of the IMF and World Bank annual meetings — how the world's most powerful institutions gather in a city that floods every year to announce that the future has arrived, and that it arrives with a bill attached.
I had spent the morning, as I spend most mornings, reading the older ledger first. The baht had been drifting against the dollar in a band that told me more about global risk appetite than any crypto chart, and the dollar index had been grinding higher in the way it does when capital flees the periphery for the center. That is the macro-liquidity context in which every technology story now sits: a world where the cost of money has been high for three years, where the marginal dollar is expensive, and where anything that cannot justify itself against a rising discount rate is quietly being repriced. Keep that backdrop in mind, because it is the strongest rival explanation for everything that follows.
This year the bill, according to a cluster of research that surfaced around the meetings, is being handed to the young. Stanford and ADP data show employment among workers aged 22 to 25 in AI-exposed occupations falling roughly twelve percent, while their peers in low-exposure roles gained about seven. Above the age of 31, the effect nearly vanishes. I have spent sixteen years watching the ledger breathe beneath the noise, and I cannot recall a quieter, more consequential entry — a number so small in absolute terms and so large in what it implies about the contract between generations.
I write this from inside a bear market, which changes the temperature of the argument. In a bull market, a story about displaced labor is a story about opportunity elsewhere. In a bear market — when funding has dried up, when protocols are quietly bleeding liquidity, when the only question that matters is survival — the same story becomes a story about who absorbs the loss. That is the frame I want to hold.
The Map, Not the Headline
To understand why this matters, and why it matters differently for those of us who live in the crypto ledger, it helps to lay out the map rather than the headline. Four independent sources are converging. The New York Fed reports that as of June 2026 the unemployment rate for recent graduates sits near 5.7 percent against roughly 4.1 percent for the workforce at large. That inversion — graduates faring worse than the average worker — is not new; it appeared in 2018 and widened to about 1.8 percentage points by April 2025, the largest gap since 1990. The US Census Bureau adds that starting salaries in AI-exposed majors have fallen about thirteen percent, with roughly half of that decline attributable to graduates shifting into lower-wage industries. And the IMF, through the voice of its managing director, has framed the issue in terms of timing: the young should worry first.
The distribution is the message. A shock that lands only on 22-to-25-year-olds, while sparing everyone above 31, is not a demand shock — it is a structural edit to the first rung of the ladder. And the mechanism is subtler than the headline suggests. The decline is coming from reduced hiring, not from layoffs. Firms are not firing their junior analysts; they are declining to replace the capacity those analysts used to provide, filling the gap with tools. Economists call this marginal substitution. On a spreadsheet it looks like efficiency. On a career, it looks like a door that no longer opens.
The macro backdrop deserves a full sentence of its own, because it is the thing the technology story wants you to skip. The tightening that began in 2022 did not merely raise the cost of borrowing; it repriced the entire ladder of expectations on which young workers rely. When capital is cheap, firms hire ahead of demand and train juniors they may not yet need — the young are the option premium a growth economy pays for the future. When capital is expensive, that premium is the first thing cut. Every hiring freeze I have watched in this bear market, across crypto and traditional finance alike, is the same decision made at different scales: the future is discounted more heavily, and the young are the future made concrete.
Here is the detail the headlines flatten: the new jobs being created in the same period are concentrated in data centers and low-wage services. The physical substrate of artificial intelligence — the racks, the cooling, the cabling — is labor-intensive, but the labor it demands is low-skill and low-paid. So the ledger balances in a grotesque way: the machine removes a high-wage cognitive role and adds, at best, a fraction of a low-wage physical one. The net count of jobs may hold. The net quality of work does not. If you want the one-line version: the economy is trading white-collar futures for blue-collar present tense, and booking the difference as productivity.
What the Ledger of Youth Actually Shows
I want to be precise about the evidence, because precision is the only defense against panic. The strongest form of the "AI did this" argument is the age stratification itself. If the decline were simply a recession, it would hit all age bands in exposed occupations. It does not. It hits the youngest cohort and spares the oldest, which is what you would expect if the binding constraint were the automation of tasks that juniors used to perform. That pattern is genuinely hard to explain away with aggregate-demand stories.
But the pattern is not proof, and the honesty of the researchers is the tell. The Stanford team behind the headline number describes its own conclusion as descriptive — a correlation, not a causal finding. That admission is buried in the coverage, and it is the most important sentence in the entire body of research. It means we have a phenomenon that is confirmed and a cause that is hypothesized. Those are different things, and confusing them is how policy gets made badly.
Consider the alternative explanation, which is not fringe but obvious. The window from November 2022 — when generative AI entered public consciousness — to 2024 coincides almost exactly with the most aggressive monetary tightening cycle in a generation. The occupations classified as "high AI exposure" — software, finance, professional services — are the same occupations that are interest-rate-sensitive, that hire heavily through campus pipelines, and that had over-hired during the pandemic boom. Every one of those variables could produce the exact same data shape. Rate shocks, post-pandemic hiring corrections, the technology layoff cycle, and offshore outsourcing all remain live hypotheses, and none of them is mentioned in the warning. This is the macro-liquidity primacy I keep returning to: before you attribute a labor-market wound to a new technology, you check the price of money. For three years, the price of money was punishing exactly the cohort the study blames on AI.
There is a deeper measurement problem too. "AI exposure" is not "AI adoption." The exposure scores come from evaluating whether a task could in principle be automated, not from observing whether firms actually deployed the tools. When you sort workers by exposure, you may only be sorting them by something else entirely — by how rate-sensitive, how credential-heavy, and how cyclical their jobs happen to be. Volatility is just truth seeking equilibrium, but a distorted map will find an equilibrium that does not exist. The difference between a task that can be automated and a job that has been automated is the difference between a weather forecast and a flood.
If you want the cross-industry shape, it is worth sketching in prose rather than a matrix, because matrices flatter certainty. Software development shows the earliest and clearest signal: high substitution on routine coding, high augmentation on architecture and review, already visible in the data. Customer service and administrative work carry the highest substitution rates, because their tasks are the most decomposable and the most verifiable — which is precisely what current systems do well. Finance and professional services sit one to two years behind, with the starting-salary decline as the leading indicator. Content creation is already feeling it, though the study does not say so directly. Healthcare and law lag because their regulatory walls are high, not because their tasks are immune. And the data center and logistics roles — the ones actually being created — carry low substitution risk precisely because they are physical, low-skill, and cheap. That is the shape of the transition: not a cliff, but a slope that tilts downward for the credentialed and sideways for the manual.
The Confounding Variable No One Wants to Name
I have seen this movie before, and I have seen it from the inside. In 2017, as a junior quantitative analyst at a Bangkok hedge fund, I spent months mapping the correlation between ICO capital flows and Thai baht liquidity injections rather than chasing tokenomics spreadsheets. I wrote a forty-page internal memo — "The Illusion of Decentralized Liquidity" — predicting that unregulated issuance would eventually trigger capital controls. My team ignored it. The memo was right about the mechanism and wrong about the timing, which is the most common way to be right. But the exercise taught me the habit I have never abandoned: when a new asset class claims to be creating value, first check what it is really doing to the liquidity of the system that hosts it.
What I did not see clearly then was the generational structure of the thing. The ICO boom was marketed as an entry ladder — a way for the young, the unbanked, the excluded to climb. What it actually did was extract capital from the young and concentrate it in the hands of early insiders, exchanges, and venture funds. DeFi Summer, three years later, repeated the trick with better branding. I was a risk modeler at a Singaporean protocol during that cycle, and I watched total value locked climb while the health of the underlying stablecoins quietly deteriorated. My team stress-tested the protocol's exposure to algorithmic stablecoins and published a warning about systemic fragility. It cost me my job. It also made my reputation, which tells you something about how the industry rewards the people who say the quiet part aloud. The lesson was not that DeFi was fraudulent; it was that a democratizing story and a concentrating mechanism can coexist in the same protocol, and the story usually wins the marketing while the mechanism wins the distribution.
By 2021 I had moved to studying NFTs — not floor prices, but the ethnography of belonging. I interviewed founders across three major DAOs and found that the communities that survived used tokens as membership badges rather than speculative assets. The ones that treated tokens as exit liquidity died. That insight became an essay on tokenized belonging that regulators read more closely than my colleagues did, which is its own kind of warning about where the real demand for this technology sits.
And then the winter came. The 2022 collapse of FTX was not, to me, a financial failure. It was a moral one — the failure of a custodian who forgot that the ledger is a promise. I withdrew from public discourse for a year and audited that collapse with my old graduate mentors, not as accountants but as philosophers. What we concluded is simple and brutal: the crypto industry promised to rebuild the entry ladder for a generation that had been locked out of the old one, and it built a casino with a ledger attached. The promise was real. The container was not. We minted souls but forgot the container — and that phrase has haunted every technology story I have written since.
Which brings me back to AI, and to the reason this is not a crypto story pretending to be a macro story. The structural mechanism is identical. A technology arrives with a democratizing story — access for all, tools that flatten hierarchy — and the returns concentrate in capital while the costs distribute to labor, and especially to young labor. The venture firms that dismissed the employment data, that called the panic unsupported, are the same category of actor that dismissed the stablecoin warnings in 2020. Their portfolios benefit from the acceleration, and their incentives point toward the soothing narrative. I do not say they are lying. I say they are interested, and interest is a bias even when the speaker is sincere. An investor who owns the shovel has a view about whether the ground should be dug; that view is not data.
The crypto industry should be paying closer attention than anyone, because it is the canary. When the entry ladder breaks, the first casualty is not the unemployment rate — it is trust. In the 2022 winter, the erosion was visible in on-chain data long before it was visible in prices: users pulled liquidity from custodial venues weeks before the headlines, and the protocol remembers what the user forgets. The same lag applies here. Youth unemployment will not spike; the shock is hiding in hiring freezes, which no headline tracks. The damage will show up first as a slow withdrawal from institutions — from employers, from credential systems, from the promise that study leads to stability. Between the code and the conscience lies the gap, and the gap is where generations get lost.
The Reflexive Mirror
Now the contrarian turn, because the evidence demands it. The most important thing about this debate is that it is a narrative competition, not a data dispute. On one side stands the warning coalition — the IMF, sovereign leaders, parts of the academy — arguing that intervention must come early. On the other stands the reassurance coalition, led most visibly by the venture firm a16z, whose partner publicly questioned whether AI is taking jobs at all. Both sides are looking at the same Stanford and ADP numbers. One reads them as a smoking gun. The other reads them as a rate-cycle artifact.
This is where reflexivity enters, and it is the piece almost no one names. The outcome of the narrative contest is self-fulfilling. If the warning wins, regulation tightens, deployment slows, and the short-term shock softens — which then appears to falsify the warning. If the reassurance wins, deployment accelerates, the shock becomes visible, and the warning is vindicated. The narrative does not describe the future; it selects it. Anyone who tells you they know which side wins is selling certainty they do not possess.
The warning side is not only institutional. Singapore's president, speaking as a head of state rather than an economist, framed the risk in a way that no data point can capture: the erosion of trust in the future. That is a political warning, not an economic one, and it should be read as such. When a generation stops believing that study leads to stability, the consequence is not a number in a labor report; it is a withdrawal from the institutions that hold a society together — from employers, from credentialing, from the social contract itself. The IMF measures the wound in percentage points. The president measures it in legitimacy.
And the blind spot is this: the debate is framed entirely around whether AI is guilty, and almost not at all around who should pay the transition cost. The language of "young people should worry first" is diagnostic; it is not distributive. The gains accrue to capital — to the model vendors, the compute owners, the application layer — while the costs land on the cohort that has the least capacity to absorb them. If that asymmetry is real, then the honest remedy is not a hiring subsidy that may pay for jobs that would have existed anyway, nor a hiring freeze dressed as prudence. The honest remedy is a conversation about redistribution that both the labor-friendly institutions and the pro-innovation media find inconvenient to host. The silence on that question is a loud statement.
I should also name a second blind spot, one specific to my own field. The regulation that is coming is aimed at the wrong target. The EU AI Act's high-risk category covers AI used to screen and evaluate candidates — the tool in the hiring process — not the AI that displaces the job itself. That is a category error with real consequences: it will regulate the instrument of selection while leaving the instrument of displacement untouched. The crypto industry spent a decade watching regulators chase the wrong layer — exchanges instead of the custody model, tokens instead of the incentive design — and it paid for that confusion in lost years and lost trust. I would rather not watch it happen twice.
What to Watch, and What to Rebuild
So what do we actually do with this, sitting in a bear market, watching our own ledgers bleed?
First, we hold the distinction between a confirmed phenomenon and a hypothesized cause. The phenomenon — young workers in exposed occupations losing ground — is real, cross-verified, and deserves attention. The cause — that AI alone did it — is unproven, and the confounding variables are large. The right posture is neither denial nor panic. It is verification. Until someone runs the difference-in-differences or finds a credible instrument to separate the rate cycle from the automation cycle, we are reading tea leaves with good penmanship.
Second, we track the signals that actually move first. Watch the monthly gap between graduate and overall unemployment at the New York Fed. Watch whether the Stanford and ADP researchers publish a follow-up that controls for the rate cycle. Watch whether "youth employment support" enters the formal agenda of the IMF and World Bank process, or remains a speech. Watch whether any major economy legislates an AI tax or a hiring incentive — that is the moment the distribution question becomes real. And watch the enrollment data in AI-exposed majors, because the feedback loop between education and the labor market runs on a one-year lag and tells you what the next cohort believes about its own future.
Third — and this is the part I care about most — we should be honest that rebuilding the entry ladder is an institutional problem, not a technological one. The ladder was not built by markets; it was built by deliberate structures: apprenticeships, first jobs that forgave incompetence, credential systems that promised a return. If AI is genuinely removing the rung, then the answer is not to wish it back, nor to subsidize its ghost. The answer is to build a new one — a curriculum and a workplace where young workers are trained to direct the machine rather than compete with it. That is the work I am doing now, at the intersection of central bank digital currency design and zero-knowledge privacy, trying to model how sovereign infrastructure can serve people without surveilling them. It is slow, unglamorous, and it will not trend. It is also the only version of this story with an ending that is not simply a transfer of loss from the old to the young.
The monsoon will recede from Silom in a few weeks. The delegations will fly home with their communiqués, and the water will drain, and the city will look exactly as it did before they came. But the ledger has recorded something, and the ledger does not forget. The rung is broken, and the young are the ones who notice first — which means the young are also the ones who will decide what gets built in its place.
The question is not whether AI is taking the jobs. The question is whether we will rebuild the rung before a generation decides the ladder was never real.
