Titian (Tiziano Vecellio), “Sisyphus,” 1548–1549. Oil on canvas, 237 × 216 cm. Museo Nacional del Prado, Madrid.

One must imagine Sisyphus happy.

In Titian’s Sisyphus, the man is mid-push. Muscles taut, boulder enormous above him. You know how it ends. He knows how it ends. The painting hangs in the Prado and tourists glance at it between Velázquez and Goya, which is its own kind of Sisyphean joke.

Expected value. Probability of winning times the payout, minus probability of losing times the cost. Positive, you play. Negative, you walk. For most of modern history the safe path had the highest EV. Degree, job, career. High probability, modest payout, a life. That math has gone negative. The credential costs six figures. The job it buys lasts five years if you’re lucky. The software that replaces you costs 1% of your salary and improves every quarter.

The entire professional class is now in the position that poor people have always been in. The lottery ticket is the rational choice. Not because the odds are good. Because everything else is worse. Poor people play the lottery and get called irrational by people who have never done the math from a starting position of zero. When the expected value of every other option approaches nothing, the one-in-a-million shot is the only EV-positive play on the board.

Elena is 34. Wharton MBA. Could have done investment banking. She ran the numbers before anyone published them: in two years those desks would shrink from 35 people to 3 people with good networks…

She saw the curve. So she bet. Thiel Fellowship. San Francisco. Vertical SaaS play in AI compliance. Adderall for breakfast, peptides for dinner. She did things she’s not proud of. Things she wouldn’t tell her closest friends. You trade pieces of yourself for runway and you call it hustle.

She raised a pre-seed. The product had users. Then the metrics stopped moving. Revenue was real but flat, and in venture math flat is a death certificate. The VCs who took her to dinner started taking meetings instead. Then calls. Then a message: “Received. Will respond later.” The response never came.

Here is what nobody writes in the postmortem: Elena’s product worked. Her customers used it daily. But the thing she built got subsumed not by a competitor, but by the platform itself. The application layer collapsed into the model layer. She didn’t only lose a market she also lost a category. She’s still in San Francisco. Still building version two. She told me last week she thinks this one is better. She says that every time. I believe her every time.

Then there’s Matthew. Austin. Econ degree from UT, graduated into what NACE now calls the weakest entry-level market in a generation. Smart kid, wrong year. So he went where the edge was. Kalshi, Polymarket. Prediction markets. He was good. Not in the way people say when they mean lucky. He built models. He understood calibration. He had an information edge on political markets because he’d grown up in a swing district and could read yard signs better than any polling aggregate.

For eight months he made money. Real money. Then the tools caught up. The same models he’d been building by hand started shipping as features. The edge compressed from days to hours to nothing. His savings ran out on a Tuesday. He moved back in with his parents in Cedar Park. His mother thinks he’s “between things.” His father thinks he’s wasting his degree. He’s 23 and he’s already been obsoleted twice once by the job market, once by the market he fled to.

He’s still with his parents, actually I don’t know, I haven’t had time. I’ve barely been sleeping. I’m pushing my own boulder. Writing about the future because that’s the rock I chose, and the hill is in front of me, and I can already see the top.

The Trap

Nobody else will say it plainly: the only viable response to what is happening is asymmetric risk. Bet on yourself, bet on a company, bet on a thesis about where the world is going. Get it right and you eat. Get it wrong and you’re one of the 90%. Ninety percent of startups fail. Seventy percent fold between their second and fifth year, not in the bright agony of a launch-day crash but in the slow Tuesday-afternoon bleed of a product nobody quite needs enough. The median venture-backed founder earns less than they would have at the job they quit. These are the odds. You play them anyway. Because the alternative is a role that gets automated on a Tuesday morning while you’re refilling your coffee.

But here is the part that should keep you up at night. Elena’s product, when it worked, made her customers need fewer employees. And some kid in a dorm was using her platform together with Claude to build the thing that would replace Elena’s product. And someone else was already building the thing that would replace his. The snake eats its tail. You have to take the bet, and the bet itself feeds the thing you’re betting against.

In every previous wave of displacement, the technology that did the displacing was built by a different industry. The loom didn’t ask the weaver to build it. There was a them. AI has no them. Elena builds the analyses that train the model that makes analysts unnecessary. The lawyer generates training data for the AI that turns his job into a subscription. The engineer writes the tools that write code without engineers. Everyone sharpening the blade. Everyone under it.

The market has already priced the first wave. A trillion dollars in enterprise software market cap vanished between late January and mid-February 2026. The S&P North American Software Index fell 15% in January alone its worst month since October 2008. Atlassian lost 35% in a single week. ServiceNow dropped 11% the day it beat earnings for the ninth consecutive quarter, because beating earnings no longer matters when the market is asking whether your customers will exist in three years. Apollo cut its direct lending funds’ software exposure nearly in half over 2025. Its co-head of private equity called it “the multi-car pileup that’s about to happen on the software investing highway.” The market punished companies for not adopting AI fast enough and for adopting it, because adopting it means your product becomes less valuable. A loop with no brake.

The loop works the same way at the human level. Elena uses AI to be more productive, which pressures the person next to her to use AI, which means the company needs fewer of both of them. Matthew uses AI trading tools, which compresses the edge for everyone running the same tools. I use AI to write faster, which means every other writer does too, which means writing is worth less. Each of us individually optimizing ourselves toward collective redundancy.

Some people think they can hedge this. Short software, buy puts, bet on the crash. It feels like control. But you’re placing a bet that the economy you live in will get worse, and if you’re right, you collect money that buys less in the economy you just bet against. The trade works on a spreadsheet. It doesn’t work in a society. You cannot hedge your way out of the world you live in.

So what’s left. You move toward the constraints: the places where judgment under genuine uncertainty still matters, where physical execution can’t be prompted, where trust relationships take years to form. You build ownership, You use AI as leverage instead of competing with it. That’s the Post-it note version. It might work for 15% of us. This essay is for the 15%. But being in the 15% does not mean you make it. It means you get to play. Elena is in the 15%. Matthew was in the 15%. The house edge is still there.

What the Winning Looks Like

Say you’re in the 15%. Say the bet paid. You built the company, the revenue grew, you have money. You moved to the right neighborhood. Your portfolio is hedged. You feel in control.

Now go outside.

The person who made your coffee this morning used to be a project manager. The guy driving your Uber has a law degree. Your neighbor just missed his second mortgage payment, quietly, because he’s still dressing for work every morning even though there’s no work to dress for. None of these people did anything wrong. They followed the playbook. The playbook lied.

This is not a moral argument. I’m not saying you should feel guilty. I’m saying you should feel afraid. Forty-three percent of middle managers believe AI will make them obsolete. These are people with health insurance and opinions about school districts, lying awake doing the same math Elena did two years ago, except they don’t have a Thiel Fellowship to fall back on. Among those who feel their jobs slipping, nearly half can’t sleep. A third say their relationships are deteriorating. Elena takes Adderall every morning. She calls it productivity. Matthew drinks alone on his parents’ patio three nights a week. He calls it unwinding. Nobody is measuring what happens when the startup fails and the Adderall doesn’t stop, or when “between things” becomes “this is my life now.”

When a professional class gets structurally repriced, the society around it destabilizes. Not only because of poverty poor countries are often remarkably stable but because of the distance between what people believed they were worth and what the market now pays them. The political direction is almost irrelevant. The mechanism is mechanical: status loss at scale produces radicalization. It happened in the 1930s across a dozen countries with a dozen different ideologies. The common variable was not culture or politics. It was a credentialed class discovering that credentials had stopped paying.

Then the anger needs a target. You cannot punch an algorithm. You can’t shoot a cost curve. But you can attack someone standing near you who looks like an easy explanation. The anger always finds a body. The body is whoever is closest and least protected. This has happened enough times in enough centuries that treating it as a prediction rather than a pattern is generous.

And this time the displaced class isn’t factory workers with limited institutional access. It’s the Elenas. People with networks, credentials, and a specific understanding of how power works. When they radicalize, they don’t join a militia in the woods. They fund candidates. They build media operations. They write policy papers with footnotes. The gated community is a solution to property crime. It is not a solution to a society that has decided the contract is broken.

That’s the world you win into. Even if the bet pays, you still live there. Your hedge doesn’t cover it. No hedge does. No amount of individual winning solves a collective problem. This is the part the “bet on yourself” advice always leaves out: the self you’re betting on exists inside a society, and if the society breaks, the bet doesn’t pay regardless of the return.

The Boulder

So here is the full picture. The safe path is dead. The only EV-positive move is the bet. The bet, if it works, feeds the machine that kills the safe path for everyone else. The world that results from enough people making the bet is a world that destabilizes even for the people who won. And you take the bet anyway. Because not taking it is worse.

That is the uroboros. The snake eats its tail and you are the snake and you are the tail.

Camus wrote that we must imagine Sisyphus happy. I think he was wrong about the happiness. But he was right about the imagining. You have to construct a reason to push the boulder when you can see the top of the hill and you know what happens there. Sisyphus was not punished for failing. He was punished for being clever enough to cheat death twice and arrogant enough to think he could keep doing it. The gods gave him a task that looked like work but produced nothing. Effort without accumulation. Sound familiar? You build the product, the model learns from it, the product becomes unnecessary, you build the next one. You push the boulder and gravity is not a natural force but an API update.

Elena is still pushing. Matthew stopped. I don’t know which one of them made the right call.

But I know which question they’re both asking wrong. Elena asks: how do I win? Matthew asks: why should I play? Neither question leads anywhere, because both assume the game is individual. The boulder is not Sisyphus’s problem. The boulder is the hill’s problem. The hill is the system that makes the boulder roll back. And no amount of pushing no matter how strong, how early, how clever changes the gradient.

The answer is not a better bet. The answer is a different definition of what winning means. Not “how do I capture more value” but “what do we do when value itself gets redefined.” That’s not a Post-it note. It’s the only conversation that matters, and almost nobody is having it because we’re all too busy pushing.

I’m at my desk at 3am with my hands on the rock and the hill in front of me. I can see the top. I know what happens when I get there. The boulder will come back. I’ll walk back down. I’ll push it again.

But I’m starting to think the point isn’t to get the boulder to stay. The point is to look up from the rock long enough to notice that there are a thousand other people on a thousand other hills, all pushing, all watching it roll back, and none of them talking to each other. The boulder is a single-player game in a multiplayer problem. We keep playing it because it’s the only game anyone taught us.

It’s dumb. I know it’s dumb. So I take the bet. But I’m starting to suspect the bet isn’t enough.

· · ·

Issue 3 will ask the question this essay has been circling: what replaces the bet. Not a better strategy for winning the current game, but what happens when we stop defining value as something you capture and start defining it as something you build that doesn’t roll back down the hill.

Alejandro Soumah writes about markets, AI, and the structural repricing of everything.

Witches’ Sabbath (Goya, 1798)

You're the Overvalued Asset

Your parents made less money than you do. They owned a home, raised three kids on one income, retired with a pension, and never once described themselves as “grinding.” Your dad worked at the same company for 30 years, thought a 401(k) was a tax form, and still ended up fine.

You make more on paper. Two degrees. 50 hour weeks. Four books about compound interest on your nightstand. You just got outbid on a one bedroom by someone who bought it in cash to rent back to people like you. You did everything right and the math still doesn’t work.

Because the math was never going to work. The cost of dignity has surged, and not for the reasons people argue about on cable news. Interest rates, immigration, zoning laws, these are symptoms. The driver is simpler: Western labor costs too much for what it produces. The gap has been covered by household debt ($17.5 trillion as of Q4 2024, per the New York Fed), transfer payments, and asset price inflation for roughly 40 years. That cover is running thin.

This isn’t a policy failure. It’s a price correction. And like most corrections, the people holding the overvalued asset are the last to realize they’re long.

The asset, in this case, is their own labor.

How We Got the Premium

For most of the 20th century, a worker in Ohio and a worker in New Dehli could perform comparable tasks, but the one in Ohio earned 15 to 40x more. The standard explanation was productivity differences, capital intensity, institutional quality. Some of that was real. But a meaningful portion of the wage gap was not earned through output. It was extracted through structure.

Prebisch and Singer documented the mechanism in the 1950s. The terms of trade between commodity exporters and manufacturing economies were structurally unequal and deteriorating. The Congo mined the cobalt. California sold the batteries. Bretton Woods made this official. The Washington Consensus doubled down. For decades, Western workers captured the spread between input costs and finished goods pricing. That spread showed up as a paycheck. It was really a subsidy.

Then capital was liberalized. Trade barriers came down. And capital did what every textbook says it will: it allocated toward the highest risk-adjusted return regardless of geography. Between 1990 and 2015, approximately 800 million people in East and South Asia moved out of extreme poverty (World Bank). The global labor supply that Western workers were competing against effectively tripled.

The premium corrected. U.S. manufacturing employment fell from 17.6 million in 1990 to 12.3 million by 2016 (BLS). Real wages for non-supervisory workers were essentially flat for three decades.

The political reaction was economically incoherent. The same populations that had championed free markets for decades pivoted to tariffs and immigration restriction overnight. They weren’t defending capitalism. They were defending an arbitrage that had closed. The policy equivalent of calling your broker and demanding he reverse a trade because you don’t like the P&L.

The Second Wave

If this were just about manufacturing, the story would be over. The factories left. The Rust Belt adjusted, partially and painfully. Some workers retrained. Most didn’t. The economy moved on.

But the second wave is here. And it’s not coming for the factory floor. It’s coming for the office.

Consider Mr. Smith. Mid-office VP at a financial services firm in New Jersey. $150k all-in. He manages no capital directly. He produces reports, synthesizes information across teams, sits in meetings where his function is pattern recognition and communication. Competent. Stable. Exactly the profile that every firm has hundreds of.

Every one of those tasks is now automatable at 60 to 80% of his current quality level, at less than 1% of his cost.

This is not a forecast. The products shipped in 2025. They’re being piloted inside the same firms that employ Mr. Smith, usually by a team two levels above him that he’s never met.

Mr. Smith costs his firm roughly $350k fully loaded: salary, benefits, office space, compliance overhead, management time. A software agent performing 70% of his workflow runs $1,000 to $2,000 per month. Even layering in human review for the remaining 30%, total cost drops 60 to 75%.

No CFO looks at that number and does nothing. Not in this rate environment. Not with margins under pressure.

The pattern is already repeating across sectors:

Legal. AI contract review tools process documents at roughly 90% of associate-level accuracy. A BigLaw associate billing at $400 to $600 per hour is competing against software that costs pennies per page. Thomson Reuters reported that firms using AI review tools reduced document review staffing by 30 to 50% in 2025.

Consulting. McKinsey, BCG, and Bain have all announced “AI-augmented delivery” programs. The language is careful, but the math is simple: a case team of six becomes a case team of three with an AI layer. Entry-level analyst hiring across MBB dropped 25% YoY in the 2025 recruiting cycle.

Financial services. Bloomberg reported in January 2026 that Goldman Sachs, JPMorgan, and Morgan Stanley are all piloting AI agents for equity research, compliance workflows, and client reporting. The initial results showed 40 to 60% time savings on standard deliverables.

Mid-office operations. Compliance, reporting, reconciliation, data entry. Any function where the output is a document or a data transformation. The BLS classifies roughly 15 million U.S. jobs in these categories.

These aren’t warehouse workers or truck drivers. These are people with degrees, mortgages, 401(k)s, and the assumption that cognitive work was the safe bet. That assumption priced in a world where AI was a research project, not a deployed product with a Stripe integration and a sales team cold-calling their employer.

The Compounding Problem

Globalization repriced physical labor. Automation repriced repetitive labor. AI is repricing cognitive labor. These are not independent shocks. They compound.

The mechanism is straightforward. A company lays off 15% of its white-collar workforce and redirects the savings into AI tools. Output stays flat or improves. Margins expand. The stock rallies. The board approves the next round of cuts. Meanwhile, the displaced workers reduce spending. The businesses that served them see revenue soften. Those businesses invest in AI to protect their own margins. The cycle accelerates.

The junior analyst who spent 80 hours building a comp table. The associate attorney reviewing 10,000 documents for relevance. The consultant assembling a market sizing deck from publicly available data. These were never high-skill tasks. They were high-cost tasks done by high-cost labor because no cheaper substitute existed.

Now one does.

The speed matters. Previous waves of labor displacement unfolded over decades. The shift from agricultural to manufacturing employment took roughly 50 years. Offshoring played out over 20. AI displacement of knowledge work is moving in quarters. Indeed postings for software, finance, and consulting roles fell 28% YoY as of December 2025. The adjustment mechanisms that labor markets depend on (retraining, geographic mobility, sectoral reallocation) were designed for much slower timelines.

We are watching, in real time, the repricing of the human intelligence premium. The financial system, from mortgages to private credit to equities, was built on the assumption that this premium would persist. The question is how fast that assumption gets marked to market.

What Comes Next

The honest question for anyone reading this in 2026: is your output something a model can’t approximate at a fraction of your cost?

If you’re not sure, you’re not alone. Most people in the crosshair don’t have a framework for evaluating their own exposure. There’s no index for “probability my job function gets automated in the next 18 months.” No way to look at a screen and see how the market is pricing your obsolescence. The data exists in fragments: Indeed postings, BLS revisions, earnings call transcripts where CEOs talk about “efficiency gains” without saying who’s being made efficient.

What we can say is that the repricing has started. It’s showing up in hiring freezes before it shows up in layoffs. In flat headcount budgets while revenue grows. In job postings that stay open for 6 months and then quietly get pulled. In reorgs where three teams become one team and an API.

The adjustment will not be smooth. Previous waves of labor displacement gave workers decades to adapt. This one is moving in quarters. The institutions designed to absorb the shock, retraining programs, unemployment insurance, higher education, were all built for a world where disruption moved slowly enough for humans to get out of the way.

That world is over.

Issue 2 will examine which specific sectors and job functions face the most acute repricing risk over the next 12 to 18 months, and what the early data is already telling us.

Subscribe now