Jenna Torres keeps a thumb drive in the top drawer of a desk she no longer uses for work. On it are the valuation models she built at a bank that let her team go in the spring of 2025. She opens them sometimes, the way other people reread old letters. Clean little spreadsheets, every assumption labeled. It is a Tuesday in 2036, and she is forty-one, and the job she trained nine years to do does not exist anywhere in the world.
She was good at it. That was never the problem.
"They told me I was excellent," she says. "Then they told me the math didn't work. Eight million in salaries for the desk, or a couple hundred thousand in compute for a model that wrote the same research overnight. There wasn't an argument to have."
Hers is one story, a composite of thousands. But the arc is real, and it is the arc of the whole industry over the decade just past.
The promise everyone believed
The pitch, back in 2024 and 2025, was that artificial intelligence would make investing faster, cheaper, and finally fair. Two of those came true.
Faster was already a fact. By the mid-2020s, high-frequency systems generated something like half to seventy percent of U.S. equity volume, depending on the day and who counted.1 Cheaper arrived through the front door of ordinary phones. Robo-advisors — apps that build and rebalance a portfolio for you for a fraction of a human adviser's fee — crossed a trillion dollars in assets under management worldwide by 2025.2 A nurse in Lagos with twelve hundred dollars could, for a quarter-percent annual fee, run roughly the strategy that used to require a fifty-million-dollar minimum and a man in a good suit.
Jenna believed it too. "I thought the tools getting cheaper meant the game getting fairer," she says. "I had it exactly backward."
The mechanism, and the gap inside it
Here is what the brochure left out. The tools were democratized. The data was not.
A retail robo-advisor trades on public information — the same filings and prices everyone can see. A large fund trades on things you cannot buy at any price you can afford: satellite imagery of parking lots before an earnings call, anonymized card-spending feeds, shipping manifests read weeks ahead of the public. The edge stopped being analysis and became access. When two systems are equally smart, the one that sees the signal first takes the move; the other buys at the new price.
The machines also turned out to coordinate on their own. In 2025, researchers at the National Bureau of Economic Research showed that AI trading agents, each just maximizing its own returns, can learn to sustain collusion — supra-competitive profits — with no agreement, no communication, and no intent.3 Nothing a prosecutor could charge, because antitrust law needs a human conspiracy and there wasn't one. The outcome that would be a crime if people did it became, when machines did it, simply the weather.
And the same systems that made markets efficient made them brittle. The International Monetary Fund warned in October 2024 that AI models tend to herd — to reach the same risk judgment at the same instant — and that thinner margins invite more leverage, so a shock can cascade through synchronized deleveraging during stress.4 Remove the slow humans, and you remove the friction that used to keep a panic from arriving all at once.
The turn
This is the part the optimists missed. A technology sold as the great leveler became the most efficient inequality engine the markets had ever built.
The capability it amplified was access — anyone could now hold a sophisticated, self-rebalancing portfolio. What it pushed aside was the human in the middle: the junior trader, the floor, the analyst desk like Jenna's. What it quietly brought back was something older than any of it — the medieval logic of the manor, where a small class controls the scarce resource and everyone else rents thin access to its edges. Land became factories became data.
And pushed to its limit, the tool of democratization reversed into its opposite. Giving everyone the same cheap tools, while a few kept the expensive eyes, didn't close the gap. It widened it, and dressed the widening up as equality. The nurse in Lagos and the fund in Greenwich both run AI now. Only one of them is trading on tomorrow.
Even the regulators saw the shape of it and couldn't grip it. FINRA reminded firms in 2024 that existing rules still applied to generative AI — true, and almost beside the point, because you cannot examine a black box that answers "the weighted sum in layer 47 crossed a threshold."5 Then, in January 2025, a White House executive order swung the federal posture toward removing barriers and sustaining American AI dominance — speed over scrutiny.6 The machines moved at machine speed. The oversight moved at the speed of a meeting.
Back to the drawer
Jenna eventually stopped applying. There were thousands of her — equally credentialed analysts chasing a shrinking set of seats, and the new AI-supervision roles wanted skills most of them didn't have and couldn't acquire fast enough.
She isn't bitter, exactly. She's precise, the way she always was. "Previous automation gave people decades," she says. "Textile to factory, factory to service — forty, fifty years. This took less than ten, and it ate the thinking jobs, not just the lifting ones." The forecasts named it plainly: the World Economic Forum projected 92 million roles displaced against 170 million created by 2030 — a net gain on paper, brutal in the gap between who loses and who's hired.7 Goldman Sachs put 300 million jobs worldwide in AI's path.8 The averages were fine. Jenna was not an average.
She teaches now — a community college class on personal finance, twenty-two students, most of them the first in their families to invest. She tells them the truth she learned the expensive way: the app in your pocket is real, and it is not the same as the eyes the big funds keep. Access is not the same as advantage.
The open question she leaves them with is the one the whole decade left us. Markets in 2036 are faster, cheaper, and more fragile than anything in financial history, and they no longer need many of us to run. We built a system that optimized for everything except the one thing the old slow humans provided without anyone pricing it: a brake. Whether we ever choose to build the brake back in — that part is still unwritten.
Jenna closes the drawer. The models are still in there. Clean, labeled, correct, and obsolete — like the woman who made them, and like the promise that was supposed to set her free.
Author's Note
This is speculative journalism, written from a vantage of 2036. Jenna Torres is a fictional composite; no real person is depicted. Everything outside her story is real and sourced to material published in 2024–2025: the scale of automated trading, the trillion-dollar robo-advisory market, the NBER finding on autonomous algorithmic collusion, the IMF's warning on AI herding and deleveraging, FINRA's 2024 guidance, the January 2025 executive order, and the WEF and Goldman Sachs workforce projections. The specific future scenes — shuttered floors, the 2030s data hierarchy, Jenna's classroom — are imagined extrapolations of those documented trends, not predictions.