The $123 trillion question

This week's newsletter is a lengthier, deeper dive than my usual editions. I've been thinking a lot about the AI tools I use everyday, and more specifically, how embedded I am in the "AI bubble." Most people aren't aware of just how far we've come in a few short years. We're far past asking ChatGPT to write emails, so I wanted to gather my thoughts around where we might be headed.

The question I want to explore today: If a person's paid work can now be done with AI, and that curve keeps bending the way the people building it say it will, what happens to money?

There's a term for this, and it's been around before AI went mainstream. Calum Chace named it in 2016: the economic singularity, the point where technological unemployment (Keynes actually coined that phrase in 1930) stops being theoretical because machines can do any job "cheaper, faster and better" than the person currently doing it.

Almost my entire adult life has been organized around one trade: my labor for money. That trade is the foundation everything else sits on, the mortgage, the retirement account, the idea of a career. I want to actually think through what happens if that foundation moves.

The people building this aren't being subtle about the timeline.

In June 2025, Sam Altman wrote that "we are past the event horizon; the takeoff has started," and gave an actual year-by-year clock: 2025 for agents that do real cognitive work, 2026 for systems that can produce novel insights, 2027 for robots doing tasks in the physical world. We're currently living inside year one of that clock.

Dario Amodei, who runs Anthropic, wrote that AI could compress "50-100 years" of scientific progress into 5 to 10, and that "our current economic setup will no longer make sense." Demis Hassabis at Google DeepMind calls what's coming "radical abundance." Jensen Huang, who sells the chips all of this runs on, told a room at the Milken Institute that every job will be affected, and immediately.

I'm not repeating these because I think the timeline is guaranteed. I'm repeating them because the people saying this aren't bloggers. They're the ones deciding how fast to ship it.

So here's what I actually want to walk through today: not whether this happens, but how, in what order, and what it touches on the way down. I found eleven dominoes. Some are already falling. Some are still just economists' models. Let's walk through a potential scenario together, including how to set yourself up for success if this scenario comes to pass.

The first domino: the easier, unglamorous jobs.

Customer support went first. In February 2024, Klarna said its AI assistant was doing the equivalent work of 700 full-time agents, handling two-thirds of its customer service chats in its first month, and cutting resolution time from 11 minutes to under 2. By late 2025, Salesforce's Marc Benioff said he'd taken his own support org "from 9,000 heads to about 5,000, because I need less heads."

Junior coding followed close behind. Microsoft's Satya Nadella says AI now writes 20 to 30 percent of the code in Microsoft's own repositories. Google's Sundar Pichai gave a similar number. Anthropic's own usage data shows coding and technical work make up roughly a third of all Claude conversations, far above its actual share of the economy.

The second domino: the ladder collapses.

This is the one that should worry you more than the first. Stanford's Digital Economy Lab tracked millions of workers through actual payroll data and found that early-career workers, ages 22 to 25, in the most AI-exposed jobs saw a 16 percent relative decline in employment. Workers in the same jobs who were older and more experienced stayed flat or grew. The cut showed up in headcount, not pay.

Entry-level work has always been how a junior becomes a senior. Remove the bottom rung and you lose the next decade's experienced hires. Researchers at the Strada Institute put it plainly: "entry-level roles are becoming more like mid-level roles."

The third domino: the life that job was supposed to fund.

Recent college graduates are underemployed at close to 42 percent, the highest rate in years. The median first-time homebuyer in America is now 40 years old, an all-time high, with first-time buyers making up a record-low share of the market. Underneath the money sits something harder to graph: the Oxford economist Daniel Susskind argues AI "may erode the meaning that people get from their work. One executive told Fortune after being laid off: "I didn't just lose a job. I lost the scaffolding I'd built my professional identity on."

The fourth domino: if wages fall, who buys anything?

This is the oldest worry in the automation debate. The popular version says Henry Ford paid workers $5 a day so they could afford the cars they built. That's a myth. Ford did it to stop turnover so bad that the company hired 52,000 men in 1913 just to keep a workforce of 14,000 filled. But the actual economic argument underneath the myth is real, and it belongs to people like Nick Hanauer: "middle-class consumers, not rich businesspeople like us, [are] the true job creators." If AI collapses enough paychecks, the market for AI-made goods collapses with them.

Here's the part that makes this interesting and potentially unstoppable. A March 2026 paper called "The AI Layoff Trap" modeled why no single company can stop this even if it wants to. Each firm "captures the full cost saving from automation but bears only a fraction of the demand loss it creates in the product market, the rest falls on rivals." That turns a bad idea for the economy into a rational decision for every individual firm, which is exactly how you get into a downward spiral.

The fifth domino: power moves from your paycheck to whoever owns the machine.

Sam Altman said it back in 2021, before any of this was live: "even more power will shift from labor to capital." Economists Anton Korinek and Donghyun Suh modeled the math. Wages depend on a race between automation and capital accumulation, and "if the complexity of tasks that humans can perform is bounded and full automation is reached, then wages collapse."

There's a second casualty here. Governments fund themselves through income taxes on labor. Korinek, working with Lee Lockwood at Brookings, wrote that transformative AI "may gradually erode the two main tax bases that underpin modern tax systems: labor income and human consumption," right as demand for a safety net goes up. Bill Gates made the same point more plainly back in 2017: "you can't just give up that income tax."

The sixth domino: your mortgage.

JPMorgan's own wealth management arm has floated a genuinely new kind of risk. Not another 2008, where subprime borrowers with weak credit defaulted. This time the concern is prime borrowers, people with 780+ credit scores and 20 percent down, whose income destabilizes after the loan is already approved. To be fair to the data, the bank's own reporting adds that "that worst-case scenario isn't reflected in the data" yet. But a Redfin and Ipsos survey found 59 percent of Americans already believe AI is eliminating jobs and making housing less affordable, a number that holds across party lines.

The seventh domino: when job loss becomes an economic crisis.

Pension funds are some of the largest owners of the exact AI companies doing the displacing. A pension fund trustee named Angelo Calvello wrote this year: "we're slitting our workers' throats with their own capital." His point: "the plan's investment capital comes from workers," and it's being deployed, in part, to fund the AI systems displacing those same workers. "Fewer employed union members mean fewer contributions and a shrinking asset base."

Right now, that second half is winning. Public pensions assumed a 6.87 percent return this year and actually got closer to 9.5 percent, largely riding the same AI boom that's threatening their contributor base.

On the flip side, a popping AI bubble might buy employees some time, but these same pension funds will suffer tremendously from a rapidly declining stock market that is almost entirely dependent on the ongoing AI tech boom.

Additionally, mortgages and pensions aren't separate risks. They're the same risk wearing two hats. Pension funds don't just hold stock in companies, they hold mortgage bonds directly, as a core piece of their fixed-income portfolios. That's not new or unusual. It's actually one of the oldest, most boring parts of how a pension fund is built: steady, long-dated mortgage debt that's supposed to match steady, long-dated retiree payouts.

The problem is what happens when the mortgages inside those bonds stop performing. We already have a real precedent for this. In 2008, pension funds were among the largest institutional buyers of mortgage-backed securities and collateralized mortgage obligations, right alongside investment banks and insurers. When subprime borrowers defaulted at scale, the value of those bonds collapsed, and pension funds took the loss directly on their own balance sheets, not just through some abstract market downturn.

So if AI-driven job loss starts pushing defaults higher, even among prime borrowers who look nothing like the subprime borrowers of 2008, the transmission channel into your retirement account already exists and has already been tested. A pension fund can get hit from multiple directions at once: fewer working members contributing to the fund, an AI bubble bursting, and the mortgage debt already sitting inside the fund losing value. Your mortgage and your pension aren't two separate line items in your financial life. For the institution managing your retirement money, they can be the same bet.

The eighth domino: does UBI actually fix this?

I want to give you real data here instead of an opinion, because we happen to have some. Sam Altman personally funded a three-year study that gave 1,000 people $1,000 a month, no strings, and tracked what happened. People worked 1.3 fewer hours a week. Total household income still rose about $6,100 a year once you count the payment. Education and job training went up 14 percent in the final year. Black recipients were 26 percent more likely to have started a business by year three, and women showed a 15 percent jump in entrepreneurial activity. Early gains in mental health and stress faded by year two.

I find the fact that Altman is the same person who ran this experiment more interesting than the result itself. He wrote the essay arguing we should tax capital and redistribute ownership back in 2021. Then he actually tested a version of it instead of just publishing the idea.

But taxing capital is a big if. Right now, our tax system won't be able to support an increase in government spending to fund a UBI, especially if our elected leaders are reacting to widespread job losses, which lead to a dramatic decline in taxable income. The question is simple: do politicians have what it takes to dramatically overhaul our tax system, and can we solve this problem globally?

Alternatively, governments can print more money to fund a UBI, but we've already seen what money printing does to the economy like during the COVID-19 pandemic. Prices skyrocket as the value of currencies drop, creating a vicious cycle where a UBI provides less and less assistance over time. In fact, inflation hits non-asset holders the hardest because the value of assets increase with inflation, while everyday goods and services only get more expensive.

The ninth domino: a tale of two cities.

The top 10 percent of Americans hold 93 percent of all stocks, the highest concentration on record, based on the most recent Federal Reserve data available, from late 2023. I already discussed how the average home buyer age has been climbing rapidly. If widespread AI automation is possible, then we're at risk of creating a permanent underclass. Those that have assets are much more prepared to weather the storm, while the young or those that don't own assets like real estate become trapped in a UBI-based system where asset prices accelerate away from them faster than they can accumulate wealth.

Let's do a sanity check.

None of this has shown up cleanly in the aggregate data yet. The Budget Lab at Yale found no unusual disruption to the overall mix of American jobs, and noted that a lot of what's blamed on AI was already underway back in 2021. A Danish study of 25,000 workers found close to zero measured effect on wages or hours, even for the youngest, most AI-exposed employees.

Harvard Business Review reported that many of the layoffs companies attributed to AI in the last year were made in anticipation of what it might do, not because of what it had actually done, and some were quietly reversed. Klarna, the company I opened with, walked part of its own AI support back and rehired humans, with its CEO admitting "we went too far."

There is also an ongoing disagreement between experts. Citi's research team argues AI growth can coexist with rising unemployment and deflation at the same time, with gains flowing mostly to what they call a "small AI elite." Torsten Slok, chief economist at Apollo, argues the opposite using an old idea called the Jevons paradox: "when steam engines made coal more efficient, Britain didn't burn less coal, it burned more." His case is that cheaper AI-driven legal, accounting, and consulting services expand the market enough to create more jobs, not fewer.

So here's the fair version. The frozen hiring market explains part of the entry-level squeeze. AI explains another part. Anyone who tells you the exact ratio with total confidence is guessing. What isn't in dispute is the order. If this keeps going, it starts at the bottom, with the youngest workers, in the jobs that are easier to automate.

If labor's value is this uncertain, ignoring the uncertainty isn't a strategy. So here's what I actually think holds up, no matter which version of this future shows up.

Own something. Deliberately, not by accident.

Altman's own proposed fix for the world he's building is to tax capital instead of labor, and use it to distribute ownership back to people. Whether or not that policy ever happens, the underlying logic is worth taking seriously. If power is moving from paychecks to owners, you want to be standing on the owner's side of that line.

Not everyone agrees ownership pays off the way it sounds. The economist Noah Smith calls the "robot lords" story too simple, arguing that "if you build lots of capital, the return on capital goes down." He's probably right that owning shares in the same five companies as everyone else isn't a strategy either.

A 401(k) that happens to include the AI winners is not the same as deliberately owning equity in something you built or chose. It's part of why, when I think about what I actually want for my son Roman by the time he's twenty, it's less about which career he picks and more about what he already owns before he needs a paycheck at all.

Be the thing AI cannot copy.

The technologist Kevin Kelly wrote: "when copies are free, you need to sell things which can not be copied." Marcus Collins made the sharper, more current version of the same point this year: "the AI-dominated future doesn't belong to the creators, but instead, the curators."

Taste, trust, and a real relationship with a real audience are the one thing AI genuinely cannot manufacture. I didn't build this newsletter by being the fastest at anything. I built it by being someone 50,000 people decided to trust with their inbox every week. That's the mechanism Kelly was describing decades before AI.

Both of these survive every version of the future.

Fast or slow, utopian or brutal, the dominoes I walked through today have led me to one conclusion, and it's how I'm living my life in anticipation of an uncertain future: own something real, and be someone real.

Hit reply and tell me which domino worried you most, and did I miss anything? I read every single response.

Jenny

P.S. If someone in your life is watching their industry change shape right now, forward them this one. It's long, but it's the one I think will matter most in five years.

Shared this email? Subscribe here.

Follow me on LinkedIn | Instagram | TikTok | YouTube | Facebook