Taylor Lorenz’s Power User - The AI Job Apocalypse: Silicon Valley Is Bracing for a Permanent Underclass w/ Jasmine Sun

Episode Date: May 20, 2026

What happens if AI becomes smarter than humans at most jobs?SUPPORT MY WORK: Buy a paid subscription to my newsletter at usermag.co Support my work on Patreon for bonus episodes, monthly Q&A livestreams, and more: http://patreon.com/taylorlorenz Researchers, economists, and top AI executives are increasingly warning about the rise of a “permanent underclass”,  a future where millions of workers lose not just their jobs, but their economic value altogether. In this episode of Power User, I sit down with writer Jasmine Sun to unpack the terrifying idea that artificial intelligence could permanently reshape class, wealth, labor, and power in America.We discuss OpenAI, Anthropic, AGI, automation, white collar layoffs, Silicon Valley ideology, AI job displacement, universal basic income, data center protests, populism, and whether society is prepared for what’s coming next.Could AI destroy upward mobility? Will automation create mass unemployment? Are tech companies being honest about the future they’re building?We cover:The permanent underclass theory explainedWho is actually pushing this ideaOpenAI's shifting stance on job displacementLeading the Future and AI lobbyingWhy AI is different from past tech hypePolicy failures and democratic backlashWhat China is doing differentlyWhat workers should actually do now#AI #Tech #Technology #ArtificialIntelligence #FutureOfWork #AGI #OpenAI #Anthropic #SiliconValley #TechNews #JobAutomation #AIJobs #TechIndustry  

It's this like slightly hyperbolic fear that this underclass could grow to encompass basically every current worker in American society as soon as AI gets good enough to do all of our jobs. Are you about to lose your job and become trapped in the permanent underclass doomed to a life of servitude to AI robots and billionaires? As AI systems rapidly improve, researchers, tech executives and economists are all putting forth this idea that millions of ordinary people could lose not just their job, but their economic. value altogether. Soon, the biggest AI companies could reach artificial general intelligence and permanently shift wealth and power towards the corporations and elites who own the technology. Meanwhile, the rest of us struggle to survive. This idea is described broadly as the rise of the permanent underclass. And Jasmine Sun, a phenomenal freelance writer, recently wrote a piece for the New York Times exploring this idea. Today, she's joining me to talk about this topic,

unpack whether or not we are actually all doomed, and what we can each do to navigate this incredibly rocky future. Jasmine, welcome to Power User. Thanks so much for having me. Okay, so I want to start off by just explaining where this idea of the permanent underclass came from. You wrote this fantastic New York Times fees. Can you tell me a little bit about the origins of this concept and just the idea of an underclass of society? So I had started to see the permanent underclass meme go around AI circles and text circles like in 2025, early 2025, mainly, mostly on like Twitter and X, though I feel like I then saw it sort of appearing on YouTube, Instagram, and TikTok also. So at first I thought that maybe it originated with the AI conversation,

but then when I tried to do some research to understand when this concept really appeared, I realized that its roots were a lot deeper. And so like the first appearance of underclass that I could find in an American context was in the 1960s. And it was actually a lot, there are a lot of parallels to what we're thinking about today with AI because it was in this book by the economist Gunner Myridal, and he called Challenges to Affluence, where he's writing about urban poverty. And because of the post-war automation boom, there are all of these factories that are getting automated and all these factory workers in cities who are getting laid off. And so very similar to like a lot of the fears of automation we have today, you have classes of

people who are working, who are happy, who had very stable lives, technological change sweeps in. And through no fault of their own, these workers lose their jobs. And the thing that Mirdal writes about in his book is that they become part of an quote unquote underclass of society, not just in the sense that they've lost their jobs. Like, okay, maybe now you can get a new job. But A, it's like quite hard to retrain and to reskill into new jobs. So many of these people struggle to do so. And B, once you lose your job, you're not just, you know, economically excluded from society, but you're also often socially and politically excluded from society and end up on the margins, not really participating in civic life.

Sometimes that leads to cycles of generational poverty. And so this idea of an underclass has to do with the notion that there is a persistent, bottom, rung, or marginal side of society, people who are not only poor, but also socially and civically excluded. And that automation waves, these sort of disruptions caused by technological shocks, for example, can be one of the things that creates an underclass. And then, throughout since then, the idea has appeared in different contexts in American society. At one point, it gets racialized to talk about.

sort of, whether it's black urban communities or whether it's undocumented immigrants who people describe as an underclass because they too are maybe working low-weight jobs without actually political representation or participation in communities. And so it becomes this phrase that is a combination of economic and political disempowerment. And now with AI, I think the new part of the fear is the permanent part of the permanent underclass, where it's not just a sort of underclass that is limited to one group in society. It's not just an underclass you might cycle in and out of, but the idea that because AI and robotics will get so good,

it will be able to fully automate any human labor. So any job a person could do, AI could do too. If that's true, then everyone in society who's currently a worker, which is most of us, end up being in the underclass because we don't have the economic mobility to do our job any better than a robot can. So like for a capital owner, for a wealthy person, Why would they pay a human when they could pay a robot? And so the idea is that when AI gets really, really good, everyone will get frozen in their current class positions.

If you're a worker now, you'll never be able to earn money again because AI can do your job. If you're a rich person now, you can spend that money to employ or rent infinite amounts of robot and machine labor to get even richer and keep building your wealth. And so it's this like kind of slightly apocalyptic, slightly hyperbolic fear that we are all not just like a few people here and there, not just temporary underclasses, but this underclass could grow to encompass basically every current worker in American society as soon as AI gets good enough to do all of our jobs. Yeah, I feel like it's, and it's always sort of pushed in the context of we're on the precipice of this. Like it could come, you know, at any moment. Who are the main people putting forth this idea? And especially in the AI context, kind of what are they pushing? And why are they

kind of talking about this concept of the permanent underclass? So the interesting thing is this meme is one that appeared in the context of Silicon Valley with regards to AI. So it's a lot of young tech workers spreading the meme on social media, whether that's founders or employees at startups and AI companies who are saying, I got to escape the permanent underclass. So that's sort of the way that people talk about it is like, I got to start a startup because that's the only way I can build wealth so that I can escape the permanent underclass. I got to get a job at Open AI or Anthropic because with that startup equity, I'm going to get super rich and I'm going to escape the permanent underclass.

Like, if you are not doing these things, if you're not using AI, you are going to get stuck in the permanent underclass. And there's this idea that you have to outrun the machines by building wealth as quickly as possible with the help of AI that's mostly being spread among these young tech workers. But I think it's also something that even people who are more senior in the space, some senior AI leaders have also expressed in other words. Like I think in a very long essay that Dario Amadeh wrote, the CEO of Anthropic in January, he also writes about the fear that AI will create, he doesn't use the word permanent, but create an underclass of people of quote unquote lower intellectual ability who might not be able to, again, outcompete the AIs. And so I think there's this pervasive anxiety that I'm noticing, like, among just workers,

leaders, executives in Silicon Valley who are all thinking, okay, if this AI is going to be so smart, like what value do the rest of us have? And they're sort of like fomenting these fears that every human is going to fall behind when AI gets good enough. You mentioned Dario, but the tech companies have played a role, I would say, in kind of putting forward this vision as well. And I mean, I really started to notice this type of meme post chat GPT, I guess, 2022 on. Well, it's primarily being driven by, as you said, tech workers in Silicon Valley, who I think have this like fear and are in some cases being hit by some of these more recent layoffs as well. Like we're seeing some of the tech companies be the first to start automating tasks.

Coinbase's CEO, Brian Armstrong said like, oh, you know, we have non-technical staffers shipping production ready code immediately. So, you know, we're automating, et cetera. But it's also people that are outside Silicon Valley that are using something like chaty BT and realizing, wow, like this can do my job really easily or, you know, an analyst at a local insurance company realizing that their job can be quite easily automated away. What role does a company like Open AI play in this? And how does their kind of like corporate philosophy feed into this idea of a permanent underclass?

Yeah. So one of the things I was really interested in examining in the piece is like what is the role of these companies? Are they worried about it? Are they doing anything to stop their technology potentially being used to replace workers? Are they engaging in policy or sort of broader action? And Open AI is like a particularly interesting case because I feel like they've actually been flip-flopping a lot over, which is, you know, maybe nothing new for Open AI.

But like they have really changed their tune on this. several times over the past many years. So, like, the thing about these AI companies, like Open AI and Anthropic and DeepMine, is that they really believe they're building AGI. They really believe that their goal is to build AI that can do everything a human can do. And their mission is to create this. And I think, like, the Open AI mission defines AGI is something that can do, like, almost all economically valuable tasks, which is, basically, again, worker replacement.

And if you read, say, this blog post called Moore's Law for Everything that Sam Altman wrote in 2021, he says very, very explicitly, like at the top of his essay that, yes, AI may create all of this abundance, but unless we do something about it, it's going to be really bad for workers. It's going to really increase returns to capital, decrease returns to labor. And therefore, he even starts proposing things like taxes on capital gains very early on in 2021. And he also stands up this economic research team within Open AI over the next few years that does things like trying to understand how fast is labor substitution going to happen, what jobs are going to be affected first.

And they started publishing papers 2022, 2023, to try to measure how much job displacement might occur. But the thing is, I think one broader trend, a lot of people have noticed about Open AI is that they have these like idealistic goals, especially at the beginning. But then they sort of become more and more of this like very profit driven company that is just trying to amass as much, you know, power and wealth as possible to like fund their continued investments. And that becomes a lot less like the nonprofit research lab that they were in 2021.

and a lot more like one of these like mega corporations. And so they end up bringing in some lobbyists and policy folks, most notably Chris Lehane, who's like a real veteran of political lobbying who, you know, started one of the biggest crypto packs that, you know, lobbied for pro-crypto legislation. And he comes in and he's like, wait, why do we have these economic teams that are publishing all this stuff about how AI might displace workers? Like that makes us look so bad. Like we should not be doing this kind of unflattering research.

we should focus on being solutions-oriented and talking about the ways that AI is going to create wealth and help people. So there ends up being over the past couple years all this turnover inside Open AI, inside the economics team that I was able to really report out where a lot of the economists on their old team end up leaving the team, quitting their jobs over the last year. Because now that lobbying and policy and government affairs has gotten involved, they've sort of reduced the independence of research and started saying, we need to be telling an economic. story about how AI helps people. And most notably, one of the quotes I got from an ex-employee was, we don't want to talk about a problem until we have the solution for it. And the thing is, what that means is they'd stop talking about problems at all. And they, for a very long time, kind of shut up about any risks to the economy or to jobs that AI might pose. But then again, in the last few months, they've changed to it again. And I think because the political winds have

been shifting so much with a lot of people worried about job displacement, a lot of folks in Washington taking this more seriously. And Open AI goes, actually, maybe we do care about this. So they publish like this big white paper proposing all sorts of policy ideas to deal with, again, declining labor share potential shocks to jobs. And they start saying ideas like maybe we should have a 32 hour work week. Maybe we should have like a public wealth fund where every American gets some equity in AI companies to make sure that they can share in the wealth. And you know, they're not committing to these ideas. When I ask them like, are you guys supporting any current legislation, they were like, we're talking about it, but we're not committing yet. They have once again

started embracing these more radical economically progressive ideas and taking AI redistribution more seriously again. So Open AI, frankly, has really shifted their tune, I think, a lot, in large part driven by the PR goals that they have and whether they feel like it's more important to turn around a good story or to show as they are right now, no, actually, we are responsive to your concerns. We do take these issues seriously. So it's kind of hard to tell because I think that people don't what to think when the AI lab leaders say one thing today and like another tomorrow. Yeah. I mean, I think open AI is just so full of contradictions as well. You know, at the same time that they're sort of writing these blog posts or pushing these progressive

policies, the 32 hour work week, etc. Discussive universal income. They're also doing a lot more political lobbying. We have Greg Brockman putting an obscene amount of money into leading the future, the super PAC then and a lot of that political lobbying kind of seems opposed to those more progressive ideas. Yeah. I mean, I think one of the of the most notable examples of leading the future pack, which of course, like you said, Greg Brockman donated to and was also shaped by Chris Lehane, their government affairs had. That one of the main races that that pack was entered was like the Alex Spores in New York, right? And he's this New York State Assemblyman.

He's running for Congress. And he did two big things recently. So the first was he introduced the Rays Act in New York. And this bothered leading the future and this acceleration is so much that they spent millions of dollars on these incredibly bad faith ads, trying to crush his chances, winning a seat in Congress. And he also recently introduced an AI dividend plan that's actually, frankly, quite similar to the idea of a public wealth fund where he advocates for the idea of if we see further economic disruption, maybe we need an AI dividend that redistributes some of the wealth of

these companies to Americans through like a token tax and different mechanisms. And again, this is someone who is supporting policies not dissimilar from the ones that Open AI is saying are needed in their white paper. And yet their super PAC is the one that is spending millions of dollars trying to defeat him in the race that he actually is winning. So one of the questions that I'm really curious about is like when Open AI does talk about radical policy ideas, is it just marketing? Because they know that people are pissed off or like, are they actually going to back that up with the political might? And like, frankly, given their record, like, I'm really not sure. Yeah, I know.

I think it'll be interesting to see. I'm very skeptical of a lot of kind of these tech companies claim because as you mentioned, their marketing can shift constantly. and it's, I think it's valuable to see kind of where the money is actually flowing. You mentioned Dario earlier. And I feel like a lot of, you know, speaking of marketing, Anthropics, marketing, even about kind of surrounding AI safety, the power of these models, et cetera, has potentially fed into this idea of the permanent underclass. How is Anthropic treating this? And how do they view this issue? Yeah.

Anthropic is really interesting. Like they are weird in the sense that they're not like the way that you normally expect a tech company to do marketing, right? Like you normally expect tech companies to be more like open AI where they're like, are. product is so good. Like, what do you mean people are like depressed or what do you mean people are losing their jobs? Whereas Anthropic is actually, I think, pushing the job loss narrative probably harder than any other actor. Dario went on this media tour telling Axios, telling Davos folks, like, hey, like, there's going to be a ton of white color displacement. I think that as many as 50% of entry level white color jobs will disappear in the next few years. And we need to do something about this now

or like publishing this big essay that talks about, you know, the potential for AI to create under class. And for Anthropic, I think what they, one is I think they genuinely believe these things. I don't think it's just marketing hype. I think that when people say that, oh, like these people are just trying to hype up their own products, like having talked to these researchers, being a pretty skeptical person, like, I do think that they genuinely believe it, whether or not they're right is a different question, but they genuinely believe this is what's coming. And Anthropics sees themselves as the truth tellers. They see themselves as the industry, Cassandra, who's going to say the things that other companies won't, who is trying to build credibility by being the ones who

are honest about the risks of AI. And so they're actually publishing a lot of research about potentially bad things that AI does, which I found really interesting to contrast with Open AI where there has been all this controversy about research independence. Anthropics research teams are straight up publishing experiments that say things like, oh yeah, like a lot of people are delegating their like extremely personal decisions to Claude and it's like screwing up their life. Or a lot of workers who use Claude to help them with coding tests end up not understanding anything about the code that they just wrote. And so Anthropics actually been publishing and speaking out very regularly about AI harms at a very high level. But it also creates a kind of interesting dissonance

because, well, I think I appreciate the honesty there. And I do think that a lot of their research has been super useful. When I asked Jack Clark, they're the head of the Anthropic Institute, who leads a lot of the societal impacts work, hey, are you planning on doing any policy advocacy to back up these ideas, he basically gives like a very long non-answer. And is like, that's like a maybe thing for the future, but like we got to think about it. And again, when you look at Anthropics business model, their business model is enterprise agents, things like Claude and Cloud Co-Work that are the primary thing replacing these white-collar workers.

And so even though I think they're really well-intentioned, I'm always asking myself, like, their core business model literally depends on this job disruption. Customers buy CloudCode and CloudCowwork because these customers, want to replace their workers with AI. And so even if Anthropic is doing, like trying to warn about it, they're also, you know, earning money off the displacement. And so I think there's only so much they can really do. If you're watching this video and you like my work,

please support me on Patreon via the link below or by a paid subscription to my tech and online culture newsletter at usermag.co. That's usermag.com. I don't have any long-term brand partnerships and a lot of my content is effectively demonetized. I've lost major brand business. major brand deals for speaking out on certain issues and for challenging power.

As you can imagine, advertisers are not exactly eager to work with somebody who covers a lot of the topics that I cover and talks about the things that I talk about. These videos I make are entirely funded by you and I can't continue to make them without your support. So if you get any value out of the videos that I create and you want me to be able to create more, please support me on Patreon or Substack via the links below. On Patreon, I do bonus episodes, monthly Q&A live streams, and post frequent updates about my work. My substack newsletter gives you a bi-weekly roundup of everything that I'm seeing and reading and paying attention to online. You can also get my newsletter on Patreon.

Once again, the links to everything are below in the description. Every dollar of your support makes such a difference. It's interesting to kind of see their positioning. I feel like any time we see this race between two top companies in a space, we usually have the first mover to the market that kind of is this category definer, I would say OpenAI. And then you sort of have their most direct competitor that often tries to position themselves. as like the more responsible one, you know, I'm thinking of like Tinder versus Bumble, where Tinder was first, Bumble came second was like, no, we're, we're like the good dating app, right? But they're ultimately like kind of both feeding each other or even Uber and Lyft kind

of engaged in this type of competitiveness early on where Lyft was like, we're not like that, you know, Uber company that's that's much more irresponsible. Like we do, you know, we treat our drivers better. We care about our drivers, you know, compared to Uber, et cetera, et And at the end of the day, I mean, Open AI has a much more consumer-facing business model versus anthropic. But it seems like both of them and just the whole industry as a general, like all of these companies are interested in moving their technology forward and getting widespread adoption. Do you think this idea of the permanent underclasses is helping with that or is it scaring people? Because it seems like a lot of people, when you talk to average consumers about their negative sentiment around AI, they bring up labor concerns. And it's a reason why they are choosing not to use these products.

Yeah. I mean, it's really interesting because I think there are two ways you can go. I think as a consumer, for example, or as a normal person, like you can think either like, okay, this AI stuff seems really bad. I'm really worried about, you know, jobs. I'm really worried about these other, like I've heard this environmental stuff. Like, maybe I shouldn't use it. Maybe I should boycott it and I should resist AI in my workplace. So that's definitely one approach that we're seeing is some workers will, they'll either unionize or just personally sort of refuse to use AI because they don't want AI to spread in their workplace and put their own jobs or their colleagues jobs at risk. And that's one reaction we're seeing. But the other reaction that we're seeing is this idea of I got to escape the permanent underclass, so I better start using AI as much as possible because that's the only thing that's going to get me out. And so there's a way in which I notice that the permanent underclass kind of becomes a little bit of a self-fulfilling prophecy for some people where, you know, if you believe there's going to be a permanent underclass, then even if you don't think it's a good idea to like, say, work for an AI company or even if you don't really like using AI, you feel like it's the only way to be everybody else

and make sure you're one of the people who ends up winning or with enough money to escape automation, right? And so there's a lot of people who have this fatalism that's like there's going to be a permanent underclass, so I better make sure I'm not a part of it. Or another way in which this permanent underclass becomes a little bit of a self-fulfilling prophecy sometimes is these companies kind of assume that like eventually all the jobs are going to be automated. Like that's the natural path of technological progress. And so we just better make sure we are the ones to do the automating. And you'll see them use these benchmarks,

increasingly that measure how good AI is at replacing people's jobs. And these benchmarks aren't like risk benchmarks. These benchmarks are how they understand, are we doing valuable work? So they're almost like competing to see which AI model is the best at replacing different human jobs, whether it's like banker or real estate broker or like news writer or like whatever else. And so I think there's a bunch of ways where because people believe that the impacts are going to be so extreme and that you're going to see this really radical inequality, the way that

some people think about it is to resist. But the way that others think about it is, oh, God, I better make sure I'm on top. And I find that to be a really interesting dynamic. Yeah, definitely. I feel like there's this anxiety that a lot of people have, as you said. And it's like, yeah, you have to sort of be like clawed maxing or whatever to make sure that you don't fall behind. I mean, it's kind of like the whole like gambling economy thing too.

I mean, I, you know, I follow that a little less. But it's like with crypto and with sports gambling and like everything else. It's like, I think people feel really economically insecure right now. I think they know that like the game is rigged and like a lot of people are going to be screwed over. And I think a lot of people feel like gambling or crypto or scams or, you know, AIMaxing or whatever it is. It's their only way to get out of like a very economically precarious situation. What's interesting you mentioned crypto, et cetera. I feel like, you know, there's a lot of discussion about the tech industry overhyping, you know, new technologies and then those sort of the massive cultural or, you know, economic effects of those technologies not playing out.

Crypto is a good example of this. You know, we've seen other kind of innovations. I would even say like when they were all obsessed with the creator economy, funding, overfunding, a lot of these tools. Like Silicon Valley constantly goes through hype cycles. How much of this is just a hype cycle versus how much do you think AI is genuinely sort of game changing technology? I saw somebody on Twitter recently doing this whole thread of things that VCs have declared

the new internet. And, you know, I would say 99% of the time it doesn't shake out. AI does seem different. But, you know, as somebody that follows this industry more closely, like, where do you see it? I think I'm probably somewhere in between like AI is just like everything else we've had and the biggest AI bulls like the people in the industry who are saying it's going to replace literally all human labor, right? So like I should say clearly like I'm not a permanent underclass believer. Like I do think there will be job disruption. And I think that we should be thinking about that and policymakers should start preparing for that.

But the most extreme scenarios are probably not true. So in one sense, like I think it's notable that AI consumer adoption has been really, really fast. and firm adoption has been really, really fast. So, like, when you measure how fast a new technology diffuses the economy, like, how fast does it take for an app to get to a million users or 10 million users? I think chat chabit from that measure is, like, the fastest growing app of all time or something like that. And so there's clearly a lot of product market fit. People really want to use AI. It can do all of these things.

People find it useful or fun or whatever. And, like, it is spreading faster than, like, almost every technology that we've had before. And so we should take the speed of that seriously. I think the other thing that we should take really seriously is that AI is, to some extent, general. It can't do literally everything. Like, it has no physical presence right now. But in terms of cognitive tasks, like, it can do a lot of stuff that it's never been trained to and it's getting better pretty fast.

And so the reason I think about this in the jobs context is because with previous technologies, maybe you have a machine that can replace one task, right? And then the person will just adjust and do a different related task. But the thing about AI is, like, it can actually generalize, maybe. to the tasks around it too. So like, for example, you know, I'm a writer. And so three years ago, AI could maybe only write at the level of like a sixth grader. And so I as a, you know, professional journalist, I'm not that worried about it. Or if you're like a college student, you're probably not that worried about it. Within like two to three years, I think now AI can

write at the level of a college undergrad. And so college undergrads are now facing this threat in a way that they did not three years ago. And previously, machines and factories were not improving at that rate, where, like, they could get so much better and do so many more tasks that it's harder for humans to just readjust to something the AI can't do yet because the AI itself is just improving so quickly and improving itself. So I think that that aspect of AI is more concerning from both the jobs perspective and makes it a little different from other technologies because it's not just one task. It's so many tasks and it gets better at all of them, even faster than humans can get better at all of them. Like, AI can probably improve its own writing ability faster than

most humans can. And I think that's pretty worrying. At the same time, the reason that I tend not to be as extreme as like a lot of the people who are in the industry is I think a lot of them just don't really understand how the economy works and like haven't had enough real jobs, which like sounds like kind of silly, but I think is true. Like it is true that AI is like totally change everything about software engineering. And every software engineer right now is basically using cloud code. A lot of them don't even write code anymore. And the junior software market has totally shifted as a result. At the same time, like, software engineering is like a very specific kind of thing. Like, it's like something you can do on your computer without like talking to anybody

in person. Every part of the job is like written down into code base. So it's like very clear what all the context is. There's less like tacit knowledge that just lives in your head. It's very objective. Like either the code runs or it doesn't run. Like it's buggy or it's not buggy. So it's like super easy to test and design around. And there's all of this open source code on the internet. So you have all this training data. But like a lot of jobs in the economy, like, whether it's like, I don't know, like you work retail in a store or like you're a salesperson and you actually have to like build trust the people and have dinners and like persuade them and like learn their kids' names and stuff like that. Like there's so many jobs that have relational or physical components

or that are just more subjective and like creative and taste based that AI can't do yet. And so I think that a lot of these AI researchers when they say like, yeah, AI is going to be able to do every job. They just kind of assume that all jobs are like they. their job. They don't realize that like a lot of jobs in the economy are like super, super different and are just totally different beasts. And so they're still stuck in their mindset of like, I'm pretty smart and my friends are pretty smart. So like if it can do our job, it can do everyone's job. And it's like actually your job is like really specific and narrow and special. And so it's going to take a while before AI can do everybody else's job too. Yeah. I think about

this a lot obviously in the context of journalism where I mean, I have always had like, I think it's hard for AI to go out and like report stories. But at the same time, you know, ironically, leading the future was paying this third party consultant to run a AI generated news website. And they actually had agents emailing people requests for comments. You know, I don't know if they had them full FOIA requesting, but certainly requesting documents, et cetera. So I do think it can, it can have a little bit more agency than people think. But as you mentioned, a lot of jobs are still so person to person based. A lot of this, you know, sort of concern about the permanent underclass or this idea that the permanent underclass arriving is also to.

tied to this idea of artificial general intelligence. And it seems like that's also kind of a disputed term that some people don't even know if it's like something that can ever be reached or we don't really understand a lot about human intelligence. Yeah. So artificial general intelligence or AGI has been kind of this North Star for the industry for a lot of the past probably 15, 20 years. And the reason it originally came up is to make a distinction between artificial general intelligence

and narrow AI. So like in the 1980s, when they were building AI, all of the AI was narrow. Like, it's like we can build a chess playing bot, but it can only play chess. Like, it can never play checkers. It can never learn poker. Like, that system is so purpose built just to do one thing. And so the dream was we actually want to build systems that can do a wide range of tasks

because as we talked about before, that's like a lot more economically valuable. If you don't need to purpose, like specially train it for chess and checkers and like dominoes, it can just learn all of that at once and generalize, basically. to new tasks. And so I think that AGI was originally a useful concept because for academic researchers, they could make that distinction. And the reason that now with LLMs, people are really interested in AGI is because one of the things that happens when you train on data from all of the internet is internet contains a lot of different general tasks in it. And so chatGBT can do so many different types of things, right? Like it can write a poem and it can write code and it can generate pictures.

And like all of these are pretty general compared to the older systems. But as for the miles zone of building AGI, people totally disagree even within the industry about what that actually means. I think for some people, it is any job a human can do that is cognitive. Like, as long as it's not a physical world task, AI needs to be able to do all of those things, including, like, writing like a really brilliant novel or something like that. Another thing would be some people say that AGI is AI that can self-improve, like AI that can build the next generation of AI. And that's important to people because once AI can build the next AI, then it'll just move faster and faster and faster and you'll get what these people think about as an intelligence explosion.

You'll have definitions of AGI that are more about like crossing some economic milestone. Like AI can generate like a $10 trillion company on its own or something like this. And the thing is, like I at one point tried to make a list of every definition of AGI that like every researcher and company has ever come up with. And they just totally disagree with each other. And I think I basically concluded, like, this term doesn't actually mean anything. It is an aspiration. I think it's helpful almost as like a myth for these people to orient themselves

towards the same goal as like we want to build something that is artificial but has these like very powerful, flexible properties that can do as wide a range of tasks as well as like a human can do them that can be, you know, better than a PhD level human at every single cognitive capacity. But I don't think it's going to look like a milestone. I think that like when people, think of it, like some finish line, like, who's going to build AGI first? I don't think we're going to reach it and one day be like, okay, we all agree we build AGI.

Like if a company says that they did, everybody on Twitter is going to argue about it and be like, no, that's not AGI yet because it doesn't mean my personal definition. And so like, I feel like just arguing over, you know, what AGI is and do we have it yet is like not that useful because like AI's impacts don't arrive when we get AGI. Like AI is already part of our lives and part of the economy right now. And so when we think about like education or work or security issues or whatever, like all this stuff is popping up even before we have quote unquote AGI and it's going to continue to happen when we get AGI. So I don't want to like create that binary of like, did we build it yet?

And then only once we build it are all of these big consequences going to show up. At the same time, I think it's really important to understand the aspiration because I do feel like again as almost like a guiding myth of the industry, you can really understand these people's motivations better about what they're trying to build when you understand that they are trying to build something that is general, as flexible as a human is, and yet also smarter than humans at each one of those capacities. Is there anywhere in the economy currently that we're seeing any kind of AI-driven permanent underclass start to emerge? You know, are there any kind of economic sectors that you're seeing just get very automated successfully in a way that, you know, is quite dangerous to workers?

Yeah. So there have been some economists and economic policy folks who have tried to do research on who's the most vulnerable to AI? Because there's a bunch of things that basically determine, like, whether your job is actually vulnerable. Like, one component, of course, is, like, can AI do your job? Like, could the model, in theory, like, actually do your job? But then there's other things like, is this industry, like unionized or does it have a lot of protections so that workers can protect themselves from AI replacing their jobs, even

if it theoretically could? Or, like, public sector unions, teachers unions, or even, like, professions like doctors and nurses where they're able to use medical regulation to sort of. of mandate that, you know, a human is always in the room or something like that. That's one component. Another one is something like, how easy is it for that class of worker to retrain into something else? Because the thing is, like, if you are, you know, a highly paid software engineer and your job gets eliminated, it does suck no matter what. But it's possible that some of those folks have the savings or they have the sort of other education and skills to try and retrain

into something else where they're still useful. And so the people who are the most vulnerable to AI are people in industries that are not organized, that are not unionized, but also people who may not have as many resources to then just go learn something new. And so one category that people don't think about a lot is like clerical workers. So one of the most vulnerable categories is clerical workers who are mostly like 40 to 60 year old women and often in suburbs and rural areas who are doing kind of administrative tasks, mostly computer knowledge work tasks. Because they're a little bit older, they are not going to have a super easy time, just

learning a brand new skill from scratch once you're already older. And they're not unionized. And basically, I feel like nobody is thinking about this group of workers in, like, the media class at least or in the AI industry because they don't think about, like, you know, a 50-year-old mom in Minnesota who maybe she only is a high school education, but she's been doing this sort of clerical work to support her family for the last few decades. And if she gets laid off from her job because of AI, it's going to be super hard for her to find something else to learn and do something new from scratch.

The other category, of course, that people are talking about more is, like, college grads and, like, young people right now because we're seeing that, like, people who leave college right now are graduating into a really, really tough job market. People debate on, is it because of AI, is it not? But the fact of the matter is, like, if you are planning on doing a knowledge work job, entry level hiring is much lower than it used to be, you don't have work experience again. So, like, yes, you may have your college degree, but you don't have enough experience to, like, super easily switch into something new. You probably don't have savings. And frankly, another thing that I think people don't talk about enough is like a lot of these college students relied on AI for like all of their homework during college. And so they don't have as much of a skill base as they ought to have. And so like one thing that I'll hear from, you know, hiring managers at companies is like, why would I hire a 22 year old with Claude Code when I could just use CloudCode? Like the 22 year olds who are coming into the workforce right now literally aren't as trained as they could be because they've over relied on AI. And that makes themselves even more at risk. So I think those are two categories I think about a lot are like mostly older secretarial workers in rural and suburban areas, plus the college grads who are coming into the workforce. Already AI is shaping up to be this major issue in the 26 midterms. What has the political response from people in D.C. been to this, you know, idea of the permanent underclass. I mean, you mentioned Boris earlier who's made sort of AI a center point of his campaign, but are there people developing policies that are getting traction? How is D.C. responding? Yeah, I mean, it's really interesting. I think there's a ton of movement right now in D.C. And we're going to see more and more over the next year.

I've been spending a little bit more time moving between SF and D.C. Over the last few months, just because you can see it heating up. I think that, like, even last year, I would say that policymakers were mostly not taking AI job loss that seriously. And that's mainly because when you look at the job data, you don't really see a lot of job loss. Like maybe you see a little bit of impacts with early career workers, like these young workers. But for the most part, like unemployment seems fine.

like hiring seems fine. You don't really notice it. And policymakers, you know, Congress people are super old. They probably don't use AI themselves. Like they're just not really thinking into the future. And so they're kind of waiting for it to become a problem before they act. I think what's really changed over the last few months is that normal people, normal workers, like citizens do kind of see what's coming. Like they see their bosses talking about AI. They see the layoffs that are happening and they are getting super, super worried about what's going to happen with AI in the future. And am I going to get stuck in a permanent underclass? And that is actually putting a lot of pressure on politicians to start thinking of a plan and messaging around AI, even if it's a little

bit early in terms of the data. So I am seeing a lot of movement, both in the policy communities and in like congressional offices to start to think about these problems. I wouldn't say that most folks have already put out plans as to like, do I support a UBI? Do I support a jobs guarantee? Do I support this or that? But like, you know, I've gotten calls and emails from multiple offices, congressional offices, since writing my piece about trying to think through labor impacts, both on the right and the left. So another thing is that AI issues are really bipartisan. Like if you think about jobs, like that affects literally everybody. There's no American of any party who's not worried about their jobs. And so I think another interesting thing

is that AI is not really arranging itself across the normal partisan lines. And there are a lot of policy organizations and advocacy groups that have people from the right and the left sort of coming together to focus on issues like protecting workers or thinking about education, things like that. But again, like, you know, I don't think there's a lot of actual policies that are serious that are being put out right now. It's very much still at the ideas and brainstorming stage. And my sense is when it's really going to break out is probably the 2028 presidential

campaigns or like the gubernatorial campaigns where presidential candidates and governor candidates are going to start trying to announce really big AI plans in order to to differentiate themselves from other candidates because it's a kind of thing you can campaign on. It's like, you guys are scared about job loss. We're expecting these impacts in the next one to two years. And I'm going to be the person to guarantee, say, like, wage insurance or a jobs guarantee or like retraining programs for college grads or like whatever it is. I think that people are going to be running on flagship AI related platforms because more

and more people are starting to care about it. And it's something that, you know, candidates can differentiate themselves on. Yeah. I mean, you talk about all. also in your New York Times piece about these candidates seeking to harness this rise of populist sentiment, especially among American workers. And you also talk about how if workers hear that AI is going to entrench this permanent underclass, they will do anything to stop it. And I feel like this idea of the permanent underclass has also given rise to the ban on data center construction in places like Maine,

the fighting back against self-driving cars, these extreme sort of censorship and surveillance laws around chatbots. And, you know, I'm curious about that sort of. sort of populist rage and anger and how you view it and how you see that being channeled. I know a lot of people have also spoke about in tech that they are concerned about it or they think that this is ultimately dangerous for these politicians to be feeding into this. Yeah, it's super interesting because I mean, we saw like the guy throwing a Molotov at Sam Olman's house recently, obviously all these data center protests, like this random Indianapolis councilman who approved a data center project got like like a gunman came to his door and

shot it 13 times. Like, he wasn't hurt, but it's terrifying and left a note on the door that said no data centers in all caps. And, like, you know, maybe it's about just a data center. Maybe he's just, like, a really big nimbie who, like, really doesn't want a new building in his neighborhood. But, like, my suspicion is that with things like the data center activism, some of it is about the construction and the fact that they don't want it. I think a lot of it is actually generalized rage at AI and at corporations and at billionaires. Like, it's like we have already been, as you know very well, like in America where a lot of people, young people feel like the system doesn't work for them, feel like economic mobility is broken,

feel like billionaires are taking bigger and bigger a share of the pie, and there's, you know, you get things like support for Louis J. Mantrioni. If we don't feel like we have democratic means to stop the accumulation of wealth by the 1%, then I think a lot of very angry people start turning to whatever they can find. And for some people that's posting angry on the internet, for other people's like protesting the data center construction in their neighborhood, for other people who are very extreme. It's like assassination attempts. And I think that the problem is when people feel like they are part of a broken system

and they feel like the normal channels of you know, democratic voice don't work. They feel like voting doesn't work. They feel like normal. Just like, let me email my congressperson doesn't work. Because there's so much money in politics that there's, you know, no chance that you can do anything about it. Then they start going a little rogue and turning towards these other means

or advocating for very, very extreme policies. I mean, even with things like the data center moratorium, when I talk to folks who are in support of the moratorium, which is basically like, we are not going to build any new data centers until we come to an agreement on jobs, kids' safety, all of these different issues. Like, the politicians like Bernie and AOC who are supporting that, they know that the moratorium is not going to get passed. They don't even want to have a permanent moratorium.

What they are using is trying to find the most extreme form of leverage against the companies. And they will basically say, like, the point of the moratorium is to find one place where we have any chance of sort of intervening in what the AI company is. are doing. Any chance where there's a democratic process, which in this case is like zoning approvals or data centers or whatever, to intervene in the ability to build AI, block it there until you can come to the table with civil society, with workers, with parents, with whoever. And I think that's the thing that a lot of people feel right now is that AI is coming into everyone's lives, but there is

no democratic channel. Like, most people don't know anyone who works in AI. Most people don't know anyone who works in AI policy. If they're upset about the way that AI is showing up in their community or their school or they're just really scared about their jobs. They feel like they have no means to do anything except let me try to ban self-driving cars. Let me try to ban these data centers in my neighborhood. It's like this very localized action against a thing that is frankly not really a localized phenomenon. And so I think from my perspective, I really want to see more policy action and also communication from more policymakers at a state and a federal level to say things like, hey, like if you're like a taxi driver and Waymos show up and you lose your job to a Waymo,

because like there's no more taxi drivers anymore. Like we are going to save you. We are going to give you unemployment insurance for like two years instead of six months because we know it takes a lot of time to like train into something new. Or we are going to give you like free access to like a trade score, community college. We're going to do all these things for you. I feel like people just feel like there is no plan. There's no reassurances that they're hearing.

And if they lose their job through no fault of their own, no one's going to help them. And so of course, if you believe that, you're going to just try to ban the thing that is going to take your job. And what I really want to see is I want to see policymakers and the AI companies, frankly, responding very directly to people's concerns, saying that, like, you are right to be concerned. Like, you should have economic security and it's not your fault if, like, technology shows up and changes your job. And we are going to take some of the profits from that technology and ensure that, like, you can have health care and, like, you can have a good life and you can find time to do something new. Because that's, I think what people are responding to is, like, they don't trust the government or companies to save them right now. And like, I don't think they have a reason to trust that. You quote Alex Carp in your piece and there's a lot of rhetoric around the sort of acceleration is that are very against this, you know, idea of stopping technological progress saying, well, if we don't do this, like China is going to catch up or beat us, you know, in this sort of like theoretical AI rights.

And I'm curious, are we seeing similar concerns about a permanent underclass, like play out in other countries like China that are also adopting AI at, you know, high rates and moving forward into this like technological AI driven future. This is super interesting. Yeah. So I was in China for the past couple weeks and I did when I was talking to either AI researchers or when I was just talking to like normal people in China and my relatives like random people, I'd ask them like how do you feel about AI? Do you use it? Are you worried about jobs? Whatever. And I think it's pretty different from the US. There's a few reasons for that. So one is China has a lot of youth white color unemployment for totally unrelated reasons. And so they've been struggling with like as high as 20 or 25% youth unemployment for years. It's not because of AI. because there are too many college grads and not enough jobs for those college grads. They have lots of factory jobs, but most college grads don't want to take the factory jobs. And so they're just not doing anything. So one is like China already had a lot of high unemployment.

So AI is not really changing that. What we are seeing, though, is one on an individual level, I think China has much more of the culture of people think, I need to use AI so that I don't fall behind. So in terms of do you resist and protest or do you use AI and try to escape? I think most Chinese people lean much more towards, I need to adopt open claw, I need to use, you know, Deepseek or DoBal is the one that a lot of them use in order to make sure that I am not falling behind. Because China has always been a very economically competitive place. Like there's like, you know, over a billion people there. Every job is highly, highly competitive.

And even before AI came along, it was like, you better grind, you better work 996. You better adopt all of the technology because if you don't take this job, there are a thousand people waiting behind you to take the job. So the orientation is already towards like adopting the technology in order to win. There are other factors like I think that the Chinese government has some policies and legislation in place in order to protect employment further because unemployment is already such a big problem and the Chinese government cares so much about social stability that they are really worried about increasing unemployment at all. And so the government has added a lot of friction to prevent AI layoff.

So it's very hard for companies to lay people off. I have an aunt who works at a state-owned enterprise in China. And I asked her like, hey, do you use AI at your job? And are you worried about getting laid off because of AI? And she said, I do use AI to help me with my work. I do know that actually AI can now do the job of two people at my work. Like, I'm aware of that. But because of these legislation, like, they're not allowed to lay me off for that reason.

Like, we're all going to use AI to do our work a bit faster and easier. But like, we're not going to get laid off because there is a legal mandate that they not do that. And so, like, I think policies like that make people feel more comfortable. And I think that the U.S., if we start to see more unemployment, we should consider things like that. I think layoffs are freaking traumatic. And, like, as much as possible, if a company can instead say, let's do more with the workers that we have, rather than just laying people off, I think that is the path that people should go down. So I think that China is at least being quite a bit more proactive.

And the U.S., frankly, it's worth us watching what works and what does not work from the policies that they try to employ because they're just trying a lot more stuff. They also have more like jobs programs. Like this is again before AI, but, you know, because there's so many people and the government's really scared of unemployment because they think people are going to protest and the Chinese government's authoritarian, so they're really scared of activists. They think that if everyone has a job, they're not going to like protest or be unruly. And so you see a ton of investment in public works, like in public parks, like there's people who are sweeping clean streets in the parks. There's so many gardeners. Like I've all the immaculate landscaping.

I've never seen so many gardeners in public parks before. And they're just like really, really well-maintained public infrastructure because the government has just decided to create tons of tons of public infrastructure jobs, partly to make the public spaces nice. And, you know, it's great to have clean streets. Like, I love that. But also literally to give people jobs so that people have something to do and have some source of income, which I also think is worth us thinking about. I mean, during the works progress administration, like we also in the United States decided to just invest in our national parks and public infrastructure. And I think that's a good idea. Yeah.

I mean, I think also during the work. that same period, we saw this like expansion of just the economic safety net and which has been eroded over the sort of past several decades, which is I think part of why people are so scared of ending up in this permanent underclass, right? There's no health care. There's no system. Like, yes, healthcare is tied to employment. Like, it's like if you lose your job, how are you going to have health care? It's terrifying. So I understand kind of like American workers fears. I'm curiously like what indicators you're watching as this all plays out, you know, and what should people pay attention to it? You mentioned before people are sort of falling into this like, by

cap of like screw AI, let's kill all tech, like end it now versus like, all right, I'm going to try to adopt and become this kind of like AI enabled employee. You know, as people and listeners navigate it, like what should they be watching out for and like what like economic indicators or sort of other trend lines are you watching to see kind of which way the wind is going on this issue? Yeah. I mean, I think it's definitely worth keeping an eye on the economic data. I wouldn't focus so much on individual layoffs because it is true that sometimes companies will say we're laying people off because of AI, but it's actually for other reasons. And so those are a little bit unreliable, but like I think looking at both like unemployment data, but also like the rate of new hires.

So like one thing that we're seeing already is that layoffs are actually less common than slowed hiring. So one of the things that I expect to happen is it's not that companies will do big layoffs, but that they just won't hire new people. And so again, that affects entry level workers the most. But I would try to look at rates of new hiring for different jobs in industries. I also think that part of it is I recommend that people try to use the technology, even if you don't like it, because that's also what teaches you, what AI is good at and not good at, and teaches you whether your own job is at risk. You know, like, it's like, as a writer, like I, every once in a while,

will try to have chat, you B, T, or Claude, write my piece for me. I never copy, paste any of the sentences. I do completely the writing from scratch, but I do that because I want to know how good AI is at my job. And it's actually very helpful to sort of what I call, like, feel the jagged edges of the technology. Like one thing I learned from that is like AI is pretty good at writing a single sentence, but it's like really bad at writing like long text. Like it cannot write a 5,000 word essay or like, as you know, like it's not very good at reporting. It can't really build trust with sources.

And so I think it's worth any worker trying to use AI to see if it can do your job and doing that every like six months or something like that because that's how you're going to know the state of the technology. And you might learn like, okay, AI is really good at this part of my job. So maybe I'll actually decide to use AI to help me with that. but AI is super terrible at that part of my job. So I'm actually going to invest in that area because I think the thing that people need to be doing now, especially if you're young, is like think about the skills that are very hard for AI to develop, like whether that's relational skills, like building trust with

people, whether that's like a skilled trade or a physical thing. Like there's a lot of different skills that AI is like not that good at yet. Like entrepreneurship, art, things like that. And those might be areas that you decide to like build up in your career. Like I'm not sure if like the way to like max out and like be AI is, just to use AI as much as possible because that's where I start to get worried about, like, deskilling and, like, you overrelying on AI. And I'm not saying don't use AI. Like, I use it as part of my process. But like also, like everyone has access to those tools. I would say actually, like,

in a world where AI can do a lot, there are still some areas that it's not going to be good at for a very long time. And that's where I hope that people invest is like creativity, having good taste and judgment, like having really amazing people skills, being very empathetic, like being good with your hands and good at making stuff in the world, all of that is going to be super, super useful for quite a while. Well, Jasmine, thank you so much for joining me and chatting today. I appreciate it so much and I'm such a fan of your work. Thank you so much.

The feeling is so mutual, Taylor. All right. That's it for this week's episode of Power User. If you like my work, please, please support me on Substack by buying a paid subscription to my substack newsletter at UserMag.co. That's UserMag.com. Or you can support my work on Patreon via the link below. You can get my newsletter via Patreon.

I also publish bonus episodes, do a monthly Q&A live stream, and more. And the only way that I'm able to produce it and continue to do my freelance reporting work is thanks to the direct financial support of people like you. Truly, every single dollar of your support makes such a difference and ensures that this show can continue. So please click the links below, support my work, and I'll see you next week for a brand new episode of Power User.