The idea of a “permanent underclass” started as a meme. It’s tapping real anxieties in Silicon Valley.

A rationalist blog forum popular in the Bay Area, LessWrong is known for its sprawling, hyper-logical takes on everything from AI research to optimal reproductive strategies. But during the 2025 holiday season, a version of a post beginning with three concise and concerning sentences appeared on the site in dramatic fashion: “I turn 20 in January, and the world looks very strange. Probably, lots of jobs will be gone soon. Probably, the smartest thing alive will no longer be human.”

Those who kept reading only encountered more of the writer’s malaise: “Everywhere around me there is this incredible sense of freefall and of grasping. … I stop going to lecture; I can no longer justify the time, and no one notices in a 300 person class anyway. I spend most of my time in the research building instead. … If there are a few years left, I want to spend them fully, and this is what carries me through most days.”

The post was written by Parv Mahajan, a 20-year-old computer science undergrad at Georgia Tech. His piece is among a growing number of works explaining that AI could create a “permanent underclass,” a subcategory of (sub)humans that may very likely include you and me and everybody else we know if we do not entrench ourselves against the impending AI superintelligence, which could very well reduce the worth of all human labor to zero.

I was told that this narrative was catching on. This spring, I traveled to the San Francisco Bay Area to find out why young, bright minds like Mahajan were suddenly so pessimistic about the future — and how that anxiety was shaping the decisions that will determine the course of the rest of their lives. 

My search for answers brought me initially to the studio apartment of 28-year-old Jayden Clark. While he didn’t quite invent the permanent underclass idea, if you’ve heard of it, he’s likely the reason. “Silicon Valley’s been on top for a long time now,” he told me. “I think that comes with an inherent inability for those of us in the Valley to make fun of ourselves. We should laugh a bit more, because we’re ridiculous over here. We do some strange, strange shit.”

Clark has made a living out of exposing exactly how strange that shit really is. A former startup founder, he recently abandoned his C-suite role to become a tech-industry influencer, producing snackable social media content for those yearning to know how the sausage of Silicon Valley is really made. That content includes Swole as a Service, a YouTube channel where prominent AI researchers answer questions about the latest tech trends while lifting weights with Clark, as well as sarcastic, jargon-laden posts blasted daily to his more than 40,000 X followers, a group that includes prominent employees of major AI labs and VC firms.

Clark moved to San Francisco from Australia in 2017, not as an aspiring denizen of Silicon Valley but to study the saxophone at the city’s music conservatory. Once he arrived, he found himself impressed less by the city’s art scene and more by the promise of its tech culture. “I just found the people in the city to be far more inspiring, much more optimistic,” he said of the engineers he met around the Valley, often at the gym. He eventually swapped his saxophone for a computer keyboard and quickly built his first startup, a software company for mom-and-pop rental shops. It was around this time that Clark began sharing his most risqué opinions about the tech industry from an anonymous Twitter account.

Inside his modestly furnished apartment in San Francisco’s Pacific Heights neighborhood, an impressively broad-shouldered Clark draws my attention to just three belongings: his tucked-away saxophone, his creatine powder, and a copy of Marc Andreessen’s The Techno-Optimist Manifesto. The book is an extended essay by the billionaire venture capitalist that argues, in essence, for more tech innovation and less tech regulation, a set of ideas that comprise an ideology now known as effective accelerationism (e/acc). 

In fact, Clark himself coined the term in an anonymous 2022 Substack post co-written with three similarly pseudonymized Twitter friends. In it, they characterized e/acc as “simply the self-awareness of capitalism.” Clark has since distanced himself from the latest offshoots of the e/acc movement and rejects their more extreme “post-human” ideas, including ones that embrace the outright speciocide of humanity. But he remains staunchly pro-tech and revels in making good-natured fun of those who view AI as the future trigger of any number of possible cataclysms, including the obliteration of the job market.

The threat of a permanent underclass was thereby the perfect modern myth for Clark to lampoon in defense of technological progress. Economist Gunnar Myrdal introduced the term “underclass” to mainstream American discourse in the 1960s as a way to describe the lowest possible segment of society. Then, in 1979, the Carnegie Council on Policy Studies in Higher Education warned that the U.S. was in danger of creating “‘a permanent underclass,’ a self-perpetuating culture of poverty” if the nation did not better prepare young people for the job market. The term started appearing in the context of AI in online posts dating back to at least 2024. In the summer of 2025, it pervaded the more satirical sectors of the internet, where it caught fire as a sardonic joke. Clark, who is known online as @CreatineCycle, quickly became one of the most prolific purveyors of this mocking mind virus.

He announced on July 25, 2025, that “rather than celebrating my birthday today I am accumulating capital in order to escape the permanent underclass.” Then, a few weeks later, he advised his followers: “you have 2 years to create a podcast in order to escape the permanent underclass.”

Clark’s tweets were soon making the pages of The New Yorker alongside sincere warnings for white-collar workers. But for Clark, the permanent underclass was never anything more than a durable joke. “I think the idea of a permanent underclass is kind of ridiculous,” he told me. “I just don’t think a lot of technologists in the Bay Area understand how basic economics work.” In June, he broke character, telling his X followers, “I hate to be the bearer of good news but the permanent underclass doesn’t exist and it will all be fine.”

But finding the meme absurd hasn’t stopped Clark from spreading it. If anything, the enduring specter of the term offers a promise of career longevity for tech commentators like him. Yet the question remains whether the rest of the labor force will be in on the joke. Could one influencer’s playful parody ever become someone else’s waking nightmare?

In May, OpenAI announced that an internal model had produced a novel solution to a longstanding problem in combinatorial geometry. In a comment responding to the announcement, an anonymous mathematician wrote that the breakthrough was “existentially upsetting” and wondered, “What do we do with ourselves when our intelligence is literally unnecessary?”

Clark suggested a possible fix-all for the mathematician’s existential anxieties: “my brother in christ: welcome to the gym.”

A 30-minute drive from Clark’s apartment, in the ritzy San Francisco suburb of Hillsborough, lies a Spanish-style compound that embodies in every conceivable manner the antithesis of the permanent underclass story. Zillow tells me the 10-bedroom palace would sell for about $20 million. Other sources put the price tag closer to $68 million. 

The villa’s primary occupant is Rocky Yu, a serial entrepreneur with a knack for recognizing big technology trends early. In recent years, Yu has focused less on forging new tech products of his own and more on building a paradise where anyone with a promising idea can pursue their potentially world-disrupting digital pet project. With its arboreal pageantry of palm trees and manicured hedges in the front yard and a private koi pond out back, the 2.3-acre Mediterranean-style lot nestled between holes 4 and 12 of the Burlingame Country Club became the perfect tract of land for Yu’s tech-industry Eden.

“We take community very seriously,” Yu told me as we strolled around the mansion’s verdant front yard. “We regularly bring all the best researchers from frontier AI labs: OpenAI, Anthropic, Google, xAI. The key technology inventors come here to talk about the current state of their models.” More than anything, Yu seems proud to have built a clubhouse where some of the world’s most important people are eager to spend time. “In the early days, there were very few people training these large language models,” he said. “Most of them hung out here.”

Before Yu’s arrival, the house’s bedrooms were occupied by superstar AI developers like Andrej Karpathy, one of the founding members of OpenAI. At the time, the mansion had a reputation less for the research prowess of its occupants and more for its parties. But when Yu took over the lease in 2023, he swapped the late-night blowouts for weekly hackathons and venture capital dinners. He invited a half-dozen of the Valley’s most promising engineers to repopulate the house as a collective and use its amenities as a high-end startup garage from which they could freely tinker their way toward an AI enterprise unicorn. Guests of Yu’s have since included Google cofounder Sergey Brin and music producer Alex Pall of The Chainsmokers. Yu has also given the pad a proper name: AGI House. 

AGI stands for “artificial general intelligence,” a hypothetical strain of AI that matches or surpasses human intelligence in every measurable way. I asked Yu why he chose a name for his upscale tech dormitory that alludes to something that so many people have come to fear. “There are mainly two voices in AI. One is all pessimistic: ‘Humanity is doomed.’ And the other believes in a world of abundance,” he said. “We are strong in the second camp. I fundamentally still have faith in humanity. We have atomic bombs, but the Earth is still here.” 

This perspective was affirmed by the AI-generated wall art I found inside the house, nearly all of which depicted various visions of a techno-utopia. The people in the pictures all looked well-employed and wired-in, if not necessarily happy. 

According to Yu, AI might actually improve our relationship with work. He insists that the arrival of AI-catalyzed universal unemployment could, in fact, be an opportunity for humanity to discover some greater purpose to our otherwise toilsome way of being. “We need to find a meaning beyond this current form of living for work and working for a living,” Yu said. “But it’s a lot of work to do before we get there.”

Today, AGI House is no longer occupied by startup founders but by Yu’s team of young assistants and operators, who organize the house’s weekly community functions, most of which remain invite-only. I was shown a video of a 14-year-old boy sitting in the house’s foyer during a recent hackathon, coding up an “AI that can produce an infinite factorial.” While walking the grounds, I met two young women who had scored summer internships editing videos for AGI House’s socials. Only in this pre-singularity era might a college student be hired to edit short-form video content for a mansion that, as a bonus, they also get to live in. 

The specter of the permanent underclass has seemingly not reached this plot of land. But I wondered how long such plum gigs could last and how many of today’s young tech-industry aspirants might ever land such roles before they are automated away forever. 

After all, despite its size and splendor, AGI House has only 10 bedrooms.

Twenty miles farther south into the world’s most Promethean valley — in economic weight, if not quite physical size — lies Stanford University. The school is bordered on the north by the offices of several of the world’s most prominent VC firms, including Andreessen Horowitz and Khosla Ventures, which have collectively poured billions of dollars into AI startups in recent months. Major tech firms are offering seven-, eight-, and even nine-figure compensation packages to a handful of world-class engineers, those seemingly born to build our inevitable singularitarian future. Many are recruited right out of Stanford, the institution perhaps more responsible than any other for molding what was once known simply as the Santa Clara Valley into the Silicon Valley.

A day before more than 100 Stanford graduates walked out of Google CEO Sundar Pichai’s commencement address, I watched graduates crowding around the vibrant campus’s most photogenic corners in their red-and-black caps and gowns. (The speech they missed would conspicuously fail to mention AI or its potential impact on young professionals. Several other commencement speakers this season were booed for remarks on the subject.) Proud parents released anxious breaths held across two decades of economic ambiguity. Their children, by all conventional metrics, had made it. 

But according to Stanford sophomore Saheb Gulati, a new kind of hushed anxiety was pervading the minds of those attending one of America’s most prestigious universities. 

“I’ve had all these texts from friends who are like, ‘Uh, dude, I don’t want to be part of the permanent underclass. What do I do?’” he told me. “Often when it comes up, people go, ‘You should talk to Saheb about this.’” I asked Gulati why he, of all people, had become the sounding board for the economic anxieties of Stanford’s anxious overachievers. “I was going around talking about superintelligence,” he said. “And I have a hot take, compared to most people.”

The gist of Gulati’s hot take is that AI will be the most transformational and potentially cataclysmic technology humanity has ever built, and the stories, strategies, and wagers many past professionals have used to construct their economically stable lives may soon be rendered moot. Now it seems a growing number of Stanford students are anxiously buying into this semi-apocalyptic narrative, and Gulati has made it his personal mission to navigate his classmates through their worries. 

“Stanford is AI-curious, even AI-pilled,” Gulati told me from his tech-cluttered off-campus apartment. “I’m in this computer science class. Its nickname is ‘AI Coachella’ because we’ve had Sam Altman come in. Jensen Huang came in. Satya Nadella just came in this morning. Anybody you can think of is coming to this class and giving lectures.” If there is ever a new idea, trend, or breakthrough related to AI, it’s likely that Stanford students are among the first to hear about it. The idea of the permanent underclass was no exception. Gulati told me it was a topic of discussion on campus long before it hit the average American’s X algorithm.

This is also perhaps why the notion of a permanent underclass has provoked existential angst among members of Stanford’s meritocratic elite. “[It] runs against the entire self-concept of a Stanford student — your entire life has been built upon this idea that you are the best and you’re going to be part of the American elite,” Gulati told me. “If somebody tells you, ‘No, you are getting knocked off the top of the status totem pole,’ it’s very taxing on your psyche, and you come up with a different narrative that gives you some kind of agency.”

The agentic narrative that has caught fire among a growing number of Stanford students is that they must either join or build an AI startup to capture whatever pieces remain of the pre-AGI economic pie. “All of this energy around AI is getting channeled into the question: ‘How do I build a startup that can disrupt a new vertical using AI so I can be relevant?’” Gulati said. But this kind of self-centered response from his classmates seems disastrously myopic to Gulati — he sees AI as a challenge his generation will need to address as a collective: “For me, it’s like, ‘Holy cow, you’re trying to develop a technology that will change the way the world has functioned for the last few thousand years.’” 

In response to this growing existential frenzy among his classmates, Gulati recently helped build Plan for AI, an online platform for students and graduates trying to navigate what could be the biggest economic disruption in the history of human labor. “There’s this crazy juxtaposition here at Stanford where you either get this really extreme career advice or you go to your career center and people act like you’ll still have 40 years of employment,” he said. The most extreme end of that advice was encouraging a growing number of Gulati’s classmates to drop out and found their own AI startups. “There was just this gaping hole where there should be another vision of what to do.”

Sneha Revanur, another Stanford student, was Plan for AI’s primary developer. Chosen by Time magazine as one of 2023’s 100 most influential people in AI, Revanur rose to prominence as a youth activist by leading efforts to pass AI safety bills in California and New York. She told me she is most concerned about the outright existential risks of AI — the chance that it could escape our control and lead to species-wide extinction — but in the relatively humdrum anxieties surrounding job displacement, Revanur sees a useful lever to catalyze the broader public into more urgent political action around AI. 

“Even though I care a lot about catastrophic risk, I also recognize that job loss is the issue that is most likely to drive people to the polls around AI,” she said. “I actually think that there are lots of ways in which mass unemployment is a catastrophic risk in and of itself, where you can imagine a future where people don’t have jobs, and, as a result, democracy entirely breaks down.”

Both Revanur and Gulati say that fellow Stanford students have been comforted by the information on their website. “The coolest thing about the entire project is not the actual specific advice that we are giving people, even though I think that’s pretty cool as well,” Revanur said. “It’s taking people who have this broad anxiety and giving them a clearer picture that, even in these uncertain times, there are some things that we can feel reasonably confident about.” 

Gulati was a little more ambivalent when speaking about his own prospects. “I always wanted to have a career that would accomplish something good and useful in the world,” he said. “And I’m confronted with this reality that I might only have a few years before the value of my labor just starts to drop.” 

When we spoke, Gulati did not yet have any plans for his summer and said he was struggling to find something truly worth his time — time that he now sees as a rapidly diminishing resource. Not having a summer internship locked down by early June would be unthinkable for the traditional Stanford student, but AI is upending Gulati’s career calculus: “I think it’s forcing me to take a lot more risky bets than I would otherwise.”

I used to live in San Francisco and still have friends in the city I like to catch up with whenever I’m in town. Among them is my former roommate, who wished to remain anonymous for this piece. He is a young tech worker who entered the job market during that brief moment in history when software engineering seemed to eclipse medicine and law as the safest and most lucrative industry a young professional could choose for themselves. Like seemingly everyone else in the Bay Area, he worked for a Big Tech company, one that produces an app that you probably hate having on your phone but also can’t bring yourself to delete.

My roommate used to dislike discussing his work. But when we met in June, he was excited to share his latest career update. “Two of my friends left their software engineer jobs to start something a few months ago,” he proudly announced. “And I followed suit.” Specifically, they had all left their jobs to build their own AI startups.

I told my friend I was happy for him. Then I asked him why he would do such a thing. “AI has made building things very easy,” he explained. “What used to take months or years to build is now cut to two days, if not hours.” 

So, AI has dramatically cut the time and opportunity costs of starting an AI company. But still, why leave a sure thing for a risky and unproven endeavor? 

“You get to a point of financial security, and then you have time to think about, ‘What is my place in society? Where can I provide value?’” he said. “And maybe the answer to that question isn’t being a software engineer for a Big Tech company where you’re not doing much good. Maybe it feels better psychologically to work on your own thing.”

He also mentioned, however, that he and many other engineers in his circle had begun to see the writing on the wall when it came to AI’s ability to do their jobs. “Software engineering feels done,” he said. “If you stay in your job, you’re kind of a sitting duck. There might not be a bridge off that. If you take the exit route now, you might be able to find a solution that works before everyone gets laid off, and there are 200,000 software engineers on the market. You might be able to get ahead of that curve, make your money, and not have to worry about any of this.”

In May, a post written by Deedy Das, a well-known Bay Area VC partner, went viral among tech workers. “The vibes in SF feel pretty frenetic right now. The divide in outcomes is the worst I’ve ever seen,” the post begins. “Over the last 5yrs, a group of 10k people … have hit retirement wealth of well above $20 million. Everyone outside that group feels like they can work their well-paying (but <$500k) job for their whole life and never get there.”

I reached out to Das for comment by text, but he did not respond. I did, however, get my friend’s perspective. “I think that’s bang on,” he said. “Middle managers are in a very, very tight position right now. Most that I talk to are very nervous.” For many, a gamble on their own startup is the only way to stay competitive and not get left behind in the Bay Area dust. 

Perhaps, my friend admitted, he and many others would have landed on the startup track even if it weren’t for AI. “Growing up in Silicon Valley, I’ve always wanted to have a company and build my own thing,” he said. “But with all these new developments, it seems like it’s now or never. You’re seeing a lot more people willing to take that risk.”

Joan Williams knows what it can mean when young men start taking more risks. As a professor at UC Law San Francisco, she studies the habits of the economically precarious, and in the rise of the permanent underclass discourse, she sees something both familiar and frightening: “AI is perhaps in the process of doing to white-collar workers what automation and outsourcing did to blue-collar workers a generation or two ago. And they see it coming.”

Even if AI does not provoke the kind of catastrophic levels of unemployment suggested by whispers of a permanent underclass, Williams told me that even milder economic disruptions concentrated among certain volatile populations could just as easily ripple into social chaos. The fact that software developers, among the workers most vulnerable to AI-fueled labor upheaval, are also disproportionately male and younger than the broader U.S. workforce adds another layer of fuel to the social tinderbox. 

“Impotence goes up with unemployment, that’s how deeply men feel this,” Williams said. “For the tech bros that are now losing their jobs, this hits them where it hurts because they not only have to deal with unemployment in a very practical way, they also have to deal with unemployment in a mythic way, which is much more painful.”

Midwesterner Paul (a pseudonym) is currently feeling that pain. He has worked as a software engineer for over 20 years. He used to be employed by one of the world’s biggest tech companies. He even remembers the techno-optimistic exuberance that preceded the bursting of the dot-com bubble in 2000. “I was in San Francisco, and there were job ads in front of the movies saying that their signing bonus would lease you a BMW,” he told me. “I’m like, ‘Nothing can go wrong.’ Well, everything goes wrong.”

Paul’s post-college opportunities dried up when that first bubble burst. He couldn’t find much temp work and had to move back home. It took him a while to find another coding job, but he was eventually hired by the major tech company.

Now those days in the early 2000s don’t look nearly so bad to Paul. He told me the difficulty of the post-dot-com bubble years pales in comparison to the economic quagmire he finds himself in today. He was working his steady Big Tech job well into the early years of the modern AI boom when one day his boss put a last-minute meeting with an HR rep on his calendar. “And I’m like, ‘Oh, I know what this means,’” Paul said of his layoff. 

At first, Paul didn’t fret about unemployment. Engineers like him were supposedly still in high demand. “I did what anybody does: I started hitting my network. I started hitting LinkedIn. I started applying for jobs,” he said. “And it was just radio silence.” That silence has now lasted more than 18 months. In that time, Paul has not received a single paycheck, and his sense of self-worth has been crushed. For Paul, the permanent underclass feels less like a silly online meme and more like his new lot in life. “I’m probably the underclass now,” he lamented. “If you have a steady job, headlines about layoffs are disturbing. If you’ve been laid off and can’t find work, headlines about layoffs are a perpetual waking nightmare.”

Within the context of his own layoff, Paul believes AI to be less his executioner and more the newest convenient corporate scapegoat. “If you have layoffs and an executive says, ‘Well, it’s because we’re cheap,’ then you’re going to get crucified in the press,” he explained. “But if you say, ‘It’s forces beyond our control. We just had to do it,’ then it doesn’t matter.”

The precise reasons for his layoff are ultimately immaterial to Paul. As his family’s only breadwinner, the reverberations of his unemployment have rippled deep into his home life. “Even my wife and I, it’s not smooth sailing since I’ve been out of a job,” he said. “I’m around her more now than I was before. She and I are kind of butting heads over it, and she’s worried because of course she’s worried, right? I mean, the depression, the increase in suicide rates, domestic abuse — the list of knock-on effects [of unemployment] is huge.”

Paul told me there is only so much additional pressure and disappointment he believes he can withstand before he is pushed to do something drastic. He mentioned his health insurance was now denying him proper coverage and that he was preparing to sue his home state in response. The governor’s office was ignoring his calls and emails, but Paul felt confident they would not ignore his subpoena.

“This has been the worst period of my life — sometimes I feel like I’m testing the limits of how much terrible stuff can happen to a guy in the U.S.,” he told me. “If you are a student of history, and you want to start looking at corollaries, what happens when particularly male unemployment gets very high? It’s bad everywhere throughout history when that happens.”

Williams happened to make the exact same point when we spoke. “If white-collar workers’ futures are gutted by AI in the same way that the blue-collar workers’ futures were gutted in the 1980s and 1990s, watch out, in terms of the future for democracy,” she warned. “There’s nothing so dangerous as a man without a future.”

Nick Bostrom writes about the future, and his writings on technology have proven to be uncannily prescient, making him one of the Bay Area’s favorite philosophers despite his never actually living there. 

The former Oxford professor popularized the modern simulation argument with his 2003 paper “Are you living in a computer simulation?” His 2014 book, Superintelligence: Paths, Dangers, Strategies, then helped draw the boundaries of today’s conversations around AI years before large language models reached the cultural mainstream. The book gave genuine academic credence to the AI doom narrative and has been recommended by Elon Musk and Bill Gates, as well as countless existential risk researchers.

It might, therefore, seem discrepant that Bostrom’s latest book, Deep Utopia: Life and Meaning in a Solved World, deals with an entirely different set of questions: What happens if intelligent machines give us everything we could ever want? What might there be left to do with our lives in an AI-delivered utopia? 

Bostrom acknowledges that the transition to a utopia may not necessarily be entirely smooth. “An old world is coming to an end, accomplishments are being ripped away, and hard-earned skills rendered pointless,” Bostrom told me. “It might feel a bit apocalyptic.” He admits there could even be some serious social growing pains from this epochal evolution. “If economic support structures have not yet been put in place, this will cause widespread anxiety and huge political pressures for some kind of fix,” he said. “White-collar workers might see unemployment while blue-collar workers in some sectors see wages rise — offending the sense of entitlement that people with university degrees have. People with children in school may feel a pit in their stomach when they consider that all that homework and encouragement will be for naught.”

That said, Bostrom expects those anxieties could easily fade if AI-generated prosperity becomes accessible to all. He has been bracing for this outcome most of his professional life. “I have long held the view — since the ‘90s, if not before — that the age of human intellectual contribution is on a countdown timer,” he said. He isn’t even much bothered by the prospect of his own intellectual irrelevance: “I suppose I’ve had some time to adjust to what now appears a somewhat imminent prospect.” 

At the same time, he warned me that our mortal woes will not simply evaporate in an AI utopia. After eons of toil and struggle, humans might not go so quietly into a life without work. Bostrom quoted early-20th-century economist John Maynard Keynes to reinforce his point: “There is no country and no people, I think, who can look forward to the age of leisure and of abundance without a dread. For we have been trained too long to strive and not to enjoy.”  

According to Bostrom, our inability to “chill out” is among the greatest hurdles now standing between humanity and his utopian vision for the future. “Many countries have developed a culture that valorizes activity and work — perhaps because it is good for the economy if people are working rather than slacking off,” he said. “People have internalized these values to the point that their self-worth is based on a sense of being useful.”

Solutions to this existential challenge may require us to rebuild our lives around an entirely new concept of self-worth, one that exists independent of our ability to do economically productive work. “Instead of teaching kids to sit at their desk and complete assignments so they can become productive office workers, we could imagine an education system that teaches kids how to become great at leisure,” Bostrom said. “Cultivating the art of conversation, the ability to form deep and authentic friendships, how to practice spirituality, hobbies, sports, appreciating literature, music, art, and so on.”

With so much of our modern lives still geared toward our jobs, Bostrom’s high-minded prescriptions could feel a little far-fetched. But Bostrom loves a thought experiment, and one of his greatest philosophical weapons is to imagine our present situation from a radically different perspective: “People may look back on the current time and shudder at how cramped our lives were, how brutish our sentiments, and how sad and unworthy of human dignity it was to have to sell a large part of one’s waking hours to some employer just to make a living.”

Weeks after publishing “Turning 20 while the world turns upside-down,” the blog post that catalyzed my trip to Silicon Valley, Parv Mahajan set aside his college coursework at Georgia Tech and moved to the Bay Area to conduct research for METR, a prominent AI safety nonprofit headquartered in Berkeley. 

I caught him on a Sunday evening while he was still at the METR office, dressed in a rumpled plain hoodie. I apologized for taking time out of his weekend. 

“No, it’s all good. I’m just working anyways,” he told me.

“Do you work most Sundays?”

“Yeah, I work pretty much every day.”

“Do you like it that way?”

“I think so. There’s a lot to be done.”

I asked Mahajan how someone who should be brimming with youthful optimism at 20 came to write something so fluently dispirited. “The first thing to realize, probably, is that AI progress is extremely fast,” he told me. “And it seems to me that there’s approximately no one really taking this seriously.” 

Indeed, like Stanford’s Revanur and even Bostrom, Mahajan’s AI safety work initially focused largely on minimizing the catastrophic risks of the technology. But behind his existential concerns lay something far more mundane: “I noticed, like, man, it just seems to be the case that college students feel really, really afraid and confused right now.”

The blog post, written over Mahajan’s Christmas break, was his attempt to give voice to that confusion. The piece also aired his concerns for the people who matter most to him. “I really want my sister and my friends and my family and just people in general to have the increase in prosperity we’ve seen in the last 50 years continue,” he said. “It seems like we’re not on track for that to happen.” And what about his own prospects? “I kind of expect my job to be automated by like 2032-ish, maybe sooner,” he replied. “It’s kind of hands-off from there, right?”

Mahajan seemed confident enough in his prediction that I didn’t push; perhaps I didn’t want to hear his response. But I was curious what sort of reactions he had received to his post. “There were a lot of reactions, which were like, ‘Oh, I hope you’re okay. This made me feel depressed.’” I confessed that I had the same general response. “I meant it fundamentally as a hopeful piece, but I think it came across more depressing than I was intending,” he said. 

In fact, Mahajan’s post had so dampened the mood on LessWrong that Boaz Barak, a prominent OpenAI researcher and Harvard professor, penned his own extended counter-piece. “[Barak] wrote a post called ‘You will be OK,’” Mahajan told me, “which I largely disagree with.”

Mahajan seemed to take pride in his work at METR, especially given that he left college to spend his Sundays doing it. I asked him how he thinks it might feel to live in the kind of utopia that Bostrom envisions, one where humans might not be the go-to hires for our most pressing societal tasks. 

“I talk to a lot of people that have moved from Sweden, and they say people don’t really feel a need to work there if they don’t want to,” he told me. “Some people get really fulfilled through their work, but other people just create bespoke coffee machines in their garage and collect unemployment benefits.” He sees unexpected value in that kind of relationship with work: “It leads to a culture that produces, you know, Minecraft and stuff. Like, yeah, that sounds pretty neat.” 

Notably, Bostrom is also Swedish, but what works well for Sweden and Swedish philosophers may not fly so well in the U.S. The nation’s character, from Silicon Valley to Wall Street to the American West, has been built on the virtue of labor. The worth of every American is, in many ways, calculated by their level of productivity. “Americans are weird about work,” Williams told me. “They’ve been weird about work since the 17th century. This comes from the little subgroup of Protestants who decided to wander over from England, who had this weird hangup that the way you expressed your moral worth was through your work.”

Even if the economic gains of the AI über-class are freely redistributed through some universal basic income scheme by the government, the idea that a basic income would be enough to satisfy not only the material needs of every person in the U.S. but also their labor-hardened souls may be even more outlandish than any claim about whether or not AI might take our jobs.

“This is part of the grandiosity of Silicon Valley,” Williams said, “combined with a complete lack of understanding of what most people want from life, which is not a handout from the ruling class. … Do you want your son to earn just a basic income? Is that your goal for him?”

For what it’s worth, Gulati, now in his second year at Stanford, still believes he will have the opportunity to live a professional life of purpose. If anything, that purpose may come from minimizing the disruptions, both big and small, that AI could ultimately bring forth in our lives. “There are all these questions of post-AGI governance — the fate of our health systems, education, economics, political order — and they all have essentially zero people working on them,” he told me. “It could totally be a college student who writes the first serious paper about what the democracy of the future looks like, or who joins an AI-assisted drug discovery startup, or starts a well-scoped retraining program.”

In classic American form, the specter of the permanent underclass — whether a valid societal threat or an X meme run amok — is leading some citizens to act with greater self-interest. But for others, like Gulati, it has inspired more solidarity. The permanent underclass trope gives expression to our most basic personal anxieties and social fault lines, and now that those problems are laid bare by the threat of AI, we have an opportunity to address them. 

“I think there is another narrative that gives people agency even if AI transforms society, which is to try and contribute,” Gulati told me. “I think there’ll be lots of inequality, but I think this is a problem that you solve in different ways. We should focus more on fixing and building up our institutions, and less on you personally needing to make it out of the permanent underclass.”

This article The idea of a “permanent underclass” started as a meme. It’s tapping real anxieties in Silicon Valley. is featured on Big Think.