Is AI Going to Destroy Our Lives or Not?

Many college commencement speakers are getting booed off stage for delighting in the productivity of AI in front of the audience it threatens to displace, cheering for all the money it will make them as CEOs, largely at the expense of the students who are supposed to be celebrating their achievement of finishing school.

Eric Schmidt, former CEO of Google, gave a commencement speech at the University of Arizona and told students that AI was going to touch every profession, and that the question to answer wasn't if AI would shape the world but whether they would shape AI. He was booed. Real estate owner Gloria Caulfield told graduates that AI was the next industrial revolution, and was booed. Scott Borchetta, a record-label CEO, during his booing, told graduates that "you can hear me now or you can pay me later."

The asymmetry is really the thing that is being booed here. These students have been training their entire lives for a credentialing system that is now actively being dismantled in the name of efficiency and profit. They are booing the people on stage who have benefited from the rules of a previous game, and are now lecturing them about embracing the rules of a new one. These students aren't anti-technology, in the same way that the Luddites weren't anti-technology. They are anti-technology-without-solution and anti-future-without-hope. Of course they are booing.

I did an event at LinkedIn headquarters a few weeks ago on careers in the age of AI. It was an in-person room of about 25 recent college graduates with a livestream of people of all ages. The whole point was to talk about jobs and AI and the changing world we are all trying to figure out. Naturally, the question that kept coming up was: what do I do right now?

And what does anyone do right now? What does it mean to train your whole life for something, only to have a person of power call you a "lower-value human"? How should someone think about multiple AI CEOs saying white-collar work will be "fully automated by an AI within the next 12 to 18 months," then walking back those statements? What does a career mean anymore, and how should people think about the future they worked so hard to arrive at?

The question underneath all of that is timing. When is the life that people have been told to train for supposed to start?

According to a recent Economist/YouGov poll, over half of Americans aged 18 to 29 are pessimistic about AI, and over 60% are very worried or somewhat worried about AI replacing jobs. The feelings are only deteriorating over time, with anger and anxiety becoming the dominant emotions over excitement and hope.

It's hard to know what to do because we largely don't know what's happening yet. The data around AI and the labor market is mixed, and the people telling you that they know are mostly selling you something, usually an AI product.

What is a job in the age of AI?

To level set on my direct experience with jobs and job applications: I grew up in Kentucky and went to Western Kentucky University. I didn't even know you could major in economics until I got to college, because I had never met anyone who worked in that specific field. Once I took my first economics class, I was completely hooked. I started a blog to meet more people. I had a 4.0 GPA, three majors, was a club president and founder, and a D1 athlete, primarily to try and get a job. When it came time to graduate in 2019, I applied to over 150 jobs. I got callbacks for about 5 or 6 of them. One of my options mostly existed because a recruiter noticed the blog and the initiative and gave me the chance to try, which was all I really needed. And I got the job.

In 2019, a human being read my resume. In 2026, that doesn't always happen. The hiring funnel has been almost entirely intermediated by AI on both sides. As Derek Thompson wrote last year, it's a "plexiglass wall" with both applicants and companies throwing technology at the process, neither side really seeing each other.

In recent years, entry-level hiring has been depressed. Unemployment for college graduates aged 22 to 27 is 5.6%, up sharply from 3.6% before the pandemic. Part of it is the AI confusion: an April 2026 paper finds that when AI absorbs the work juniors used to do, firms cut junior hiring, which weakens the pipeline that creates future seniors. The result is "lost cohorts" of juniors who don't get the training they need. Part of it is pandemic overhang: a study of 243 million hires across four countries finds that when you separate the effect of AI from the effect of remote work, remote work turns up as the likeliest culprit of the recent collapse in entry-level hiring. When teams went remote, the cost of supervising and training a junior went up, so firms stopped hiring them.

Both papers point to firms giving up on their juniors. That's a core rung on the economic ladder that young people are missing out on. The booing is rational.

The entire AI conversation has been organized around one question: will AI replace your job, yes or no. It's really not clear what is happening, and that's why a framework for thinking about jobs is so important.

Alex Imas has written many excellent pieces on AI and labor market displacement. He begins one piece by discussing a widely cited 2023 paper on AI exposure, which found that around 80% of workers in the US could have at least 10% of their tasks affected by large language models. Many people interpret this as "80% of jobs are at risk." Imas points out that the reading is wrong: exposure measures whether AI can do some of the tasks in a job, not whether the job entirely disappears.

Imas says the dimensionality of a job really matters, the number of distinct tasks it involves. A job with two tasks, one of which AI can do no problem, is a high-risk job. If a firm can automate half the job, mechanically it's hard to justify keeping the worker. A job with seven tasks, one of which AI can do well, is a job that probably gets more productive, because the worker has more time to do the other six tasks that can't be automated. The resulting productivity gains, the focus effect, can raise wages instead of eliminating the role.

He compares a long-haul truck driver to a consultant. A truck driver could be automated, because it's theoretically one task: going from A to B. A management consultant, a job that includes a bundle of tasks, can have some work automated, but the political work and client communication and presentation of results cannot be. Thinking of a job as a bundle of tasks is a very useful rule of thumb because it gives you a good sense of how at-risk a chosen job might be. Luis Garicano builds on the framework from the supply side, pointing out that firms buy bundles, not tasks. They pay a human to do a bunch of interconnected, context-dependent tasks, because jobs are complex.

Ernie Tedeschi at Stripe Economics uses travel agents as an example of the economic value of human expertise. In 2000, the job was exposed to automation. Booking, fare aggregation, ticketing: all of it could be automated, and travel agent headcount fell by more than 60% from the dotcom peak. But some survived, by moving up market. Travel agents now earn close to 99% of the private-sector average wage, up from 87% in 2000. Those that survived leaned on judgment, accountability, relationships, and the ability to swoop in and fix something quickly when things go wrong at 2am. Those seem to be the pillars of strength in the AI age: being able to evaluate, having some element of taste, holding accountability, building networks, and delivering a human touch in an increasingly automated world.

We don't quite understand how human most jobs are. It drifts into Rory Sutherland's Doorman Fallacy. The job of a doorman is just opening doors, right? Well, no. It's welcoming and helping and security and a place to ask questions. A job is usually a whole lot more than it seems.

Firms are also craving some humanness. Gillian Tett at the Financial Times wrote that firms are focusing on critical thinking skills and looking to hire students who majored in the humanities rather than STEM. People optimized too much for one side of intelligence, ignoring the curation of their creative skills. Creativity is enormously valuable in the age of AI.

My take: AI is sort of like freeze-dried camp food. It does a good enough job, is filling enough, but it isn't something you want to eat every day (trust me, I tried). People crave inefficient, home-cooked meals for a reason. It's a nice thing, not a certain exact number of calories. You need camp food some days. It's good that camp food exists. But we still need real food too.

Is AI even cheaper than labor?

The technology is very expensive. Many of our assumptions about AI displacement rest on the idea that it is cheaper than labor. It might be soon, but right now that assumption is just not true.

Agentic workflows, the multi-step work building toward context and revisions and judgment, are more compute intensive than a chatbot. Upcoming frontier models could be even more expensive. In April, Uber's CTO said he had to go "back to the drawing board because the budget I thought I would need is blown away already." Azeem Azhar and Hannah Petrovic cover this in a recent piece: over 70% of companies exceeded their AI budgets in 2025. This stuff is not cheap.

Some of the enormous spend is good. Brian Albrecht published a piece called "You Are Not a Horse" that walks through the AI-displaces-everyone story by tracing the dollars. If AI makes some tasks cheaper, the money saved through efficiency doesn't vanish. It gets spent somewhere else, which creates demand somewhere else, which keeps humans employed somewhere else. For the horse outcome to happen, every dollar of spending would have to land on activities with no human labor inside them, anywhere in the supply chain.

But some of the spend is bad. Companies have fired people in the name of AI efficiency, but in many cases it hasn't really made anything more efficient. Uber's COO recently said that "tokenmaxxing" was making it harder to justify AI costs within the company, because more tokens spent by engineers wasn't leading to a measurable increase in useful consumer features. Part of this is a measurement problem, and part of it is that AI compute is increasingly constrained and therefore increasingly expensive. Jobs that are expensive to automate at the moment might be safe for a while.

When you put these pieces together, the dimensionality plus the bundle strength plus the compute economics gives you a framework that is pretty useful. Jobs at risk are low dimensional with weak bundle strength and tasks that are already cheap to run on AI: customer service, back-office document processing, entry-level analyst work. If you have a high dimensional job threaded with judgment and accountability and creativity, one that might even benefit from AI spend, you're probably relatively safe. Imas says the "relational sector," the human-intensive, provenance-rich, sometimes artisanal part of the economy where the human aspect is part of the value of the good or service, will become increasingly important.

But that requires firms to think about jobs as more than a cost metric, which they don't tend to do. Axios recently reported that companies are firing employees to cover their AI bills. Starbucks is retiring an AI inventory tool because it kept hallucinating. We are likely in a rough transition period of fires and rehires as we adjust.

Are firms actually using AI?

Well:

• Some firms are using AI. Six of the largest US banks posted $47 billion in profits in Q1 2026 while shedding 15,000 jobs, and the CEOs are now openly crediting AI on earnings calls. In 2025, Brian Moynihan at Bank of America said AI was "not a threat" to his 210,000 employees. In 2026, he announced the bank had shed 1,000 jobs through "eliminating work and applying technology."
• Some firms are using humans to train the AI to eventually fire themselves. Meta laid off 8,000 employees in an AI push, assigning 7,000 workers to AI-initiative focused teams. An anonymous Meta employee, who ended up being one of those laid off, described the company as "not at all empathetic enough or human enough in how they are leading humans through this era."
• Some firms are designing their own AI. Kirkland & Ellis, the world's highest-grossing law firm, is spending $500 million to build its own AI platform, designed using information from 250 of its lawyers about how they do their jobs. The senior generation is pouring its knowledge into a system that will eventually do the work that juniors used to do to learn the job.
• Some firms are having mixed results. A Judge Business School survey of over 600 finance and AI companies found that about 80% of private sector finance groups are using AI, but only 40% reported any profit boost from it. 60% of respondents actually expect AI to increase hiring or reskilling at their firm.
• Many firms are not using it at all. Only 1 in 5 firms across the board are actually using AI. Adoption grew 70% from late 2024 to late 2025, but it's still pretty low, considering how much money companies are spending and all the fuss it makes.

Tedeschi highlights a framework around displacement: we might not see any real AI impact until we see a recession; specific jobs might be impacted rather than the whole labor market; and those specific jobs might be fine in the long run.

David Solomon, the CEO of Goldman Sachs, wrote a New York Times op-ed titled "The A.I. Job Apocalypse Is Overblown." His argument is the standard optimistic one: the economy has absorbed technological transitions before, job growth has outpaced population growth since the 1960s, and even if AI automates 25% of work hours, people will find more productive things to do. All will be well, in the long run.

The aggregate case is fine, sure, but that's not what's actually breaking people. In April, Goldman Sachs economists published a paper telling a different story. They examined four decades of individual-level data on workers displaced by technology. Workers displaced from technology-disrupted occupations take about a month longer to find a new job and suffer real earnings losses of more than 3% on reemployment. Ten years out, their earnings are still nearly 10 percentage points behind workers who were never displaced. Tech-displaced workers experience delayed homeownership and delayed household formation, with the effects amplifying in recessions. This type of displacement has always worked like this. The costs land on individuals for a very long time.

Most economists will acknowledge that technological progress can cause some adjustment problems in the short run. What is rarely noted is that the short run can be a lifetime.

Carl Benedikt Frey

The economy can absorb AI the way it absorbed previous technologies: slowly, unevenly, with new sectors appearing at the same time that the rungs of the economic ladder are getting squeezed. But jobs are only one part of the AI conversation. Wealth creation is another.

Who gets to be wealthy in the age of AI?

True wealth comes from owning shares of companies, and the US stock market is meant to be that wealth-creation opportunity. Companies go public because that's the promise we made people: tying your retirement to the stock market is going to go really great because all of the biggest companies go public and you, the general public, can participate in the upside. But companies don't go public anymore.

Deedy Das estimates that about 10,000 people, those working at Anthropic and OpenAI and Nvidia and Meta, have hit retirement-level wealth, more than $20 million, within the past five years. Everyone outside of that circle is trying to get in, but the door has closed. The most enormous wealth engine in history is increasingly confined to just a few people. OpenAI and Anthropic are both sniffing around IPOs, but both are near $1T private market valuations, which could leave little room for more upside. As Joachim Klement wrote, the IPO of these AI companies is probably nothing more than a major transfer of investment risk from the current owners to retail investors, pension funds, and others willing to buy the hype.

SpaceX just filed to go public after 24 years. They are unprofitable on $18.7 billion of revenue, losing almost $5 billion last year, and are about to IPO at a $1.5 trillion valuation. This is a company that we are asking people to build their retirement on. If you don't own equity here, your participation in the AI boom is entirely through the lens of: is this thing going to destroy my life?

Are we stuck in the permanent underclass?

Much of the age of AI is like this: privatized gains and socialized losses. It's a contributor to the discourse around the "permanent underclass," the idea that if you don't get wealthy now, you will never be successful. I spoke to a room of Stanford undergraduates a few weeks ago and that term was floated a few times. People are living in fear of the economy.

The vibecession, the marked disconnect between economic data and consumer sentiment, is back in the discourse, with new research pointing to enormous price level variability as a driver of negative sentiment. I read the 600+ comments on Annie Lowrey's Atlantic piece about it. The ones who actually engaged her question kept talking about the concept of economic security, which appears to be defined as: a bounded downside, where an illness or layoff can't erase everything you've built; a predictable floor you can plan around; a perceived link between effort and outcome still holding; and a believable path to the next thing: the job, the house, the kid.

Economic data captures a moment, but sentiment is capturing people's concern about their economic future. People can't save, so of course they aren't going to think positively about the future they can't save for. When none of those four conditions of economic security hold, the casino looks rational. Prediction markets did roughly $25 billion in volume in April alone. A recent Wall Street Journal analysis found that 70% of Polymarket bettors have lost more money than they made. The system is designed to extract from people who need a fast win and to compound for people who can afford technology to get ahead.

There are also people opting out of the system because they have economic security. While middle-class kids are being told to embrace AI to keep up, wealthy families, including many who make the AI products, are pulling their kids out of algorithmic credentialing entirely. No-screens private schools are growing, with handwritten essays, oral examinations, in-person seminars, human access. The elite version of education is becoming analog. Everyone else is stuck talking to ChatGPT. AI probably isn't going to destroy your life, but it's accelerating the closure of doors that were already closing, and the panic about that is rational.

Why does everything take so long now?

We are facing a duration problem, which ties into the economic security issue. Things are just taking longer than they used to. Career formation takes longer. Skill compounding takes longer. Family formation, home ownership, career stability: it's all getting pushed out by about five years. The thing that used to happen between 18 and 25 is now happening between 25 and 30.

The news cycle and the social media cycle and the AI hype cycle have all gotten faster, and the timescale of building a durable adult life has gotten slower, and the gap between those two things is enough to drown in. The failures we often cite as problems, the delayed home buying and the marriage data and the wage data, are functions of a longer development timescale colliding with a faster extraction timescale. That's why the casino economy is so enticing. Everything takes so long. The despair and the doom and the booing and the financial nihilism are all rational responses to being told you are behind on a schedule that quite literally cannot apply to you.

What do you do about it?

Tyler Cowen published some notes on jobs advice in the age of AI. His first principle is to look for "messy jobs," the ones hard to describe, that change by the day, with many discrete tasks. Work in biomedical, work in energy, run experiments, gather data. Go where the capex is going. I think it's terrific advice, and wanted to compile other things I've heard over the past several months.

1. Pick the manager, not the company. The training pipelines that used to build up junior employees have been more or less gutted across most white-collar industries. The single biggest variable in whether your first job teaches you anything is the person who manages you.
2. The search is the job. One warm introduction is worth fifty cold applications. If you don't have a network, the work of building one is the work.
3. Lean into creativity. Firms want good thinkers, and they are prioritizing creativity. They want people who can look at hard problems and come up with novel solutions. The goal isn't to compete with the AI, but to be the person who knows what to do when it all stops working.
4. Build something. Networks come from building a living portfolio online. Building a voice, interacting with people, and developing evidence that you can think in public is enormously valuable.
5. Be AI native, not AI proof. Students are flocking toward AI-exposed degrees, not away. Being able to help a non-tech company automate some tasks is likely one of the fastest growing jobs out there.

The despair and the doom and the booing and the financial nihilism are all rational responses to being told that you are behind on a schedule that doesn't really apply anymore. The rules are changing. Bounded downside, a predictable floor, reward for work, and a believable path to the next thing are the conditions under which a future feels possible. For a lot of people, none of that holds. Naming that clearly is the first step to fixing it.