---
title: Permanent Underclass
description: How a shitpost about podcasts became Silicon Valley's expectation of
  a coming caste system. The sixty-year history of the word, the thin evidence beneath
  the fear and the political work the story does while it circulates.
doc_version: '1.0'
last_updated: 02-Aug-26
canonical_url: https://garden.johanneskleske.com/permanent-underclass
---

This is a [[Primer|primer]] on the permanent underclass, the phrase that has organized how Silicon Valley talks about everyone else.

> Every time someone tells you what the future looks like, ask yourself: Whose present does this serve?

**tl;dr: “Permanent underclass” arrived in the AI debate as a joke on X and left it as an expectation. The word is sixty years old and comes preloaded: Gunnar Myrdal coined it in 1963 as a diagnosis of automation, and American politics then spent two decades turning it into a behavioral slur for Black poverty. What is new sits underneath, where the worry has moved from being exploited to being surplus. The evidence for that fear is thin, and the story does four kinds of political work while it runs ahead of the data. Nobody has to coordinate any of this, and the meme form protects the whole thing from refutation, because you cannot disprove a joke.**

## Table of contents

- [The two-year clock](#the-two-year-clock){: .internal-link}
- [An old word arrives pre-loaded](#an-old-word-arrives-pre-loaded){: .internal-link}
- [From exploited to expendable](#from-exploited-to-expendable){: .internal-link}
- [The evidence gap](#the-evidence-gap){: .internal-link}
- [What the narrative does](#what-the-narrative-does){: .internal-link}
- [The story survives its critics](#the-story-survives-its-critics){: .internal-link}
- [How priming closes the window](#how-priming-closes-the-window){: .internal-link}
- [The counter-move](#the-counter-move){: .internal-link}
- [Related concepts and further reading](#related-concepts-and-further-reading){: .internal-link}

## The two-year clock

A meme account on X put it best: “You have 2 years to create a podcast in order to escape the permanent underclass.”[^1]

The joke works because every part of it is doing something. There is a deadline, and it is short. There is an escape route, and it is absurdly individual, a podcast, a personal brand, a bet on being one of the people who got out. And there is a destination for those who miss the deadline, named with a term borrowed from American poverty policy, which the joke assumes you already understand. Read it twice, and the humor turns into a set of instructions.

Kyle Chayka picked the phrase up for his New Yorker column in October 2025 and described the mood around it precisely: “A sense of dread about this impending A.I. caste system has created a new urgency to get ahead while you still can.” He called the whole thing “kind of a joke, but also kind of serious.”[^2] Seven months later, Jasmine Sun took it to the front of the New York Times opinion section after more than fifty interviews in the industry, and opened with a sentence that has been quoted ever since: “Most people I know in the A.I. industry think the median person is screwed, and they have no idea what to do about it.”[^3]

That is roughly fifteen months from shitpost to op-ed to structure of feelings. A phrase can travel from irony to expectation without ever passing through evidence.

==My claim in this primer is that “permanent underclass” in its AI version is a future narrative that primes expectations and that primed expectations perform political work in the present.== The story arranges what feels settled, what feels open, what feels urgent and what feels like somebody else's problem. That work happens whether or not the future it describes ever arrives, which is why this primer leaves the forecast itself open. Whether a large share of human labor becomes economically superfluous is a question for the coming decades, and I have no privileged access to the answer. What can be examined today is the telling: who says it, in what form, and what it does to the decisions being made this year. The method here follows from that, and the section on evidence below sets out to measure a gap.

## An old word arrives pre-loaded

Gunnar Myrdal introduced “underclass” to American debate in *Challenge to Affluence* in 1963, carrying it over from Scandinavian usage. The phrase did not come from Silicon Valley, and the history it drags behind it changes what happens when someone uses it now. His definition described a growing “class of unemployed, unemployables, and underemployed” who were being “set apart from the nation at large,” and his explanation was entirely structural: automation and rising productivity were turning unemployment into a permanent feature of the economy, a condition that no longer moved with the business cycle, while demand shifted toward the educated.[^4] There is no behavioral accusation anywhere in it. The word entered American English in 1963 as a diagnosis of automation, which is exactly what it has been recruited to describe again.

Herbert Gans traced the turn. The term resurfaced in *The Public Interest* in 1973, this time defined by behavior and simultaneously functioning as code for Black poverty. By the late 1970s the adjectives “permanent” and “trapped” had been welded on. Time ran a cover in August 1977 titled “The Underclass,” subtitled with the phrase “the unreachables.” The earliest well-documented use of the exact phrase “a permanent underclass in our society” belongs to Edward Kennedy in 1978, which means the framing was never the property of one political camp.[^5] U.S. News warned about a permanent underclass in 1979, and usage climbed steeply through the mid-1980s, when the Chicago Tribune could run a headline reading “In A Nation Of Riches, A Permanent Underclass.”[^6]

Ken Auletta's 1982 book turned it into a popular typology of social dropouts. William Julius Wilson used it descriptively in *The Truly Disadvantaged* in 1987 and argued for deindustrialization and social isolation against the culture-of-poverty thesis, defending the word itself as late as 1989.[^7] The defense did not hold. Gans argued that the label bundles heterogeneous poor people into a single supposedly deviant category, reverses victim and perpetrator, works as a racial code and licenses attacks on social policy. His own sharper alternative, “undercaste,” names exclusion of a caste kind, where “underclass” still suggests a position somewhere inside a class structure. Loïc Wacquant went further in *The Invention of the “Underclass”* in 2022, describing it as a “racialized folk devil” produced jointly by scholarship, journalism and politics; a “phantom group,” a “cognitive tool of symbolic degradation and domestication.”[^8]

Wacquant's most useful line here is the one about the direction of evidence: the label tells you more about the fears of the people using it than about the people being described. Hold onto that. It is the whole method of this primer applied sixty years early.

So the term now returns to something close to Myrdal's structural meaning, and it does so carrying two decades of stigma sediment in its grain. When a venture capitalist says “permanent underclass,” the sentence sounds like macroeconomics. The word underneath still contains the Time cover, the unreachables, and the quiet assumption that the people down there are a different kind of person.

## From exploited to expendable

Underneath the meme sits a fear that is genuinely new. It differs from ordinary anxiety about job loss.

Erik Olin Wright gave the distinction the formulation I find most useful, writing in 1984 that the underclass is “economically oppressed but not consistently exploited.”[^9] Those two words carry a great deal. Exploitation is a relationship. The exploited worker is subordinated and needed, and the need is leverage: they can strike, slow down, walk out, organize, refuse. The leverage in that dependency is what the labor movement was built on. The surplus person has no such channel. Someone who is economically unnecessary gets administered, policed, means-tested and managed, and the machinery of social protest built over a century has no obvious point of attachment.

That is the real content of the dread in the AI conversation. The fear circulating in San Francisco is about whether one's labor will be wanted at all. ==The exploited worker can stop the machine. The expendable person confronts a machine that can keep running.==

A second-order effect follows. It is what makes the fear feel categorical. Most of the machinery of social protection in rich countries is bolted onto the employment relationship: health coverage, pensions, unemployment insurance, and the legal standing to bargain at all. Someone pushed outside that relationship loses the benefits and also the address through which claims get made in the first place. Redundancy would therefore read as a change of kind, while a bad labor market reads as a bad decade.

I find this the strongest part of the AI usage. It deserves to be taken seriously. It also explains why the phrase mutates so easily into caste language, as Chayka noticed when he reached for “caste system” and for the lumpenproletariat analogy. Class positions can be renegotiated because the parties need each other. A caste is a status you are assigned and cannot bargain out of. If you believe your society is about to stop needing a large share of its members, caste is a more honest description of the outcome than class, which is why Gans reached for “undercaste” long before any of this.

The move is real, the logic is coherent, and the argument is being made about a future that has not happened.

## The evidence gap

Here the story runs into the data, and what it hits is mostly silence. Anthropic's own labor market research, published in March 2026 by Massenkoff and McCrory, found no systematic rise in unemployment in AI-exposed occupations since the end of 2022. The one signal in the paper is narrow: a roughly 14 percent decline in the job-finding rate for 22-to-25-year-olds in exposed occupations, which the authors themselves describe as “just barely statistically significant.” The study contains a second finding that sits awkwardly beside the public narrative. The most exposed workers skew older, better paid and more educated, while the entry-level jobs that dominate the public warnings sit further down the exposure ranking.[^10]

The British government's assessment in January 2026 was blunter about the state of knowledge: “The available evidence does not yet provide clear answers to many of the questions that matter most for policy.”[^11] The ILO and NASK exposure index from May 2025 found roughly one in four jobs with some generative-AI exposure and then declined to convert that into a replacement forecast, concluding instead that “transformation of jobs is the most likely impact.”[^12] Andreessen Horowitz has compiled the counter-case, which has to be read against the firm's own interest: work from the NBER, the Atlanta Fed, the Census Bureau and the Yale Budget Lab points toward broad stability, with the clearest exception being the Stanford finding of a roughly 16 percent relative decline for the same 22-to-25 cohort that turns up in the Anthropic paper.[^13]

That is the empirical basis, as of this writing, for a discourse about a permanent caste system: one narrow cohort effect, found in two studies and barely significant in one of them.

I want to be careful about what this does and does not show. It does not show that the fear is baseless, and it is early. Labor market data is slow, the technology is fast, and absence of a measurable effect in 2026 tells you little about 2032. What the numbers establish is a gap. There is a small, contested, cohort-specific signal on one side and an expectation of civilizational restratification on the other. The distance between the evidence and the expectation is the space the narrative lives in.

The asymmetry in what each side has to do is the whole story. A labor-market study has to define exposure, select a comparison group, state a time period, and report uncertainty. A meme needs a number. Two years will do.

## What the narrative does

If the story is running ahead of the evidence, what is it accomplishing while it runs?

Start with what it removes from view. When a distribution of income, security and status is presented as an output of a technology, the decisions that produce that distribution stop looking like decisions. Tax policy, labor law, ownership of models and data, the design of social insurance, and the choice about who captures productivity gains: all of it recedes behind the machine, which is described as doing this to us. Sacasas named the rhetorical move the Borg Complex, the reflex of declaring resistance futile, and Mar Hicks has shown how claims of inevitability transfer power to whoever is building the thing declared inevitable.[^14] The garden note on [[Reactance and future narratives]] works through this at length, and [[Condensation of Future Imaginaries]] describes the mechanism by which contested futures harden into single obvious ones. Naturalization turns those contested choices into weather. Once the storm has a two-year arrival time, politics becomes disaster preparation. People can buy supplies or move uphill. They lose the habit of asking who controls the dam.

Everything else follows from that first move. Once the outcome is settled, the only live question left is personal position. Where will I be standing when it lands? Sun recorded a young tech worker giving the answer in the most literal possible form: “Oh, the reason I got to take this job is cuz I got to escape the permanent underclass.”[^3] Escape is the verb the story makes available. It is the verb in the meme too, and the podcast joke is funny because it is the correct joke: the proposed solution to a structural transformation of the economy is a media side hustle. Nothing in the frame suggests that the transformation might be governed, taxed, negotiated or refused collectively. Those options do not get argued down. They never enter the room, which is a more thorough form of defeat than losing an argument. The garden note [[Exit strategies as collective surrender]] traces the same pattern in the political projects of the same milieu and puts the thesis without the term: when tech executives announce that AI will eliminate most jobs, they are constructing a narrative that serves specific interests.

The clock comes next, and it follows from the escape frame. If your only available strategy is individual positioning, then timing is everything, and a head start loses value the longer it lasts. The meme says two years. Leopold Aschenbrenner puts the AGI window closing by 2027. Nate Soares advises against relying on employment. Deedy Das at Menlo Ventures went viral in May 2026 with the arithmetic of compounding wealth before the door shuts.[^15] Urgency of this kind is productive: it moves people into AI-adjacent jobs, into equity, into the sector doing the predicting. Aschenbrenner also put money behind the forecast. His hedge fund, named Situational Awareness after his 2024 AGI manifesto, ran leverage of up to 400 percent on AI infrastructure stocks and shorted software, and when the portfolio fell 67 percent in July 2026, margin calls forced a discounted sale of the equity book to Ken Griffin's Citadel and assets dropped from roughly 45 billion dollars to roughly 10. He wrote to investors that the fund “came closer to permanent capital impairment than is acceptable to us.”[^16] A prophecy that recruits its own labor force and its own capital has a considerable advantage over one that does not.

And then there is the question of whose fear this is. Read the Chayka and Sun pieces and count who appears. These are essays about people in the AI industry, worrying about their own trajectories, in a city where they are already among the winners. Sun catches the doubleness in one line: “They muse about the postwork future while pulling 80-hour workweeks.”[^3] The people the term originally described, the ones on the Time cover in 1977, appear only as a destination. Almost nobody from that category speaks. ==In these essays the permanent underclass discourse is a conversation about the poor conducted entirely among people who are afraid of becoming them.== Wacquant documented the same pattern in the 1980s discourse.

## The story survives its critics

The story is more interesting than a simple account of elites manufacturing fear, because the actors pulling on it are pulling in opposite directions.

Dario Amodei is the best-documented amplifier. In May 2025 he told Axios that AI could produce a “white-collar bloodbath,” eliminating half of entry-level office jobs within one to five years and pushing unemployment to 10 or 20 percent. Asked by CNN for the study behind the numbers, he had none to offer. He has restated the position repeatedly since, telling Bloomberg in June 2026 that the share of work affected “eventually gets close to a hundred percent.” Daron Acemoglu and Yann LeCun have both said flatly that he is wrong.[^17] Whatever else the warnings do, they position the person issuing them as the responsible adult in the room, the one who sees clearly and says so. An interest in warning settles nothing about whether the warning is right, and Amodei may well turn out to have been early. The question I am tracking here is a different one: what keeps a claim moving while the evidence for it stays this thin.

Andreessen Horowitz attacks the same expectation from the other side. David George dismissed it in May 2026 as “the lump-of-labor fallacy with updated branding,” and Fortune noted what the firm has riding on that dismissal: regulatory pressure and the mood around its own portfolio.[^18] Two positions are on offer, both interested, both working the same thin data. The narrative needs only that the story stay in circulation, and a public fight guarantees that more reliably than agreement would.

That circulation has its own economy. The clearest statement of it came from Kwalia in May 2026, describing the discourse as half an economic argument and half a “sales pitch,” wrapped inside an “attention economy that profits from the urgency, keeping the panic urgent and easy to sell.” The same piece also lands the hardest logical objection I have found: “The ‘permanent’ framing is rhetorical: it presumes today's institutions will hold while the technology that destabilises them runs to completion, and both can't be true at once.”[^19] A technology powerful enough to make a class of people permanently surplus is powerful enough to break the institutions that would have to keep them there. Permanence is the least defensible word in the phrase and the one doing the most work. The City Journal supplied the mirror-image argument a couple of days earlier, warning that the panic itself, once converted into basic-income politics, could manufacture the dependent class everyone claims to fear.[^20] Kwalia says the phrase collapses under its own logic, the City Journal says the fear will build what it fears, and the expectation survives both, because a contested phrase circulates better than an agreed one.

Alex Karp, quoted by Sun, gives the elite's own version of the fear with unusual candor: “the country could blow up politically, and none of us are going to make any money when the country blows up.”[^21] Stated as naked self-interest, the sentence still concedes that the outcome would be bad for the people currently doing best out of the arrangement, which is a strange thing to say about a future you are helping to bring about.

Which brings us back to the joke. Sun, who wrote the essay that took the term national, says openly that “the permanent underclass is definitely in a lot of ways a meme and a hyperbole,” and that she would not claim it as a sincere majority belief in Silicon Valley.[^22] I believe her, and I think that admission makes the phenomenon worse. A claim you can be held to has to survive contact with evidence. A claim delivered as a bit passes the expectation along and disowns the assertion in the same breath. Chayka's formula, “kind of a joke, but also kind of serious,” describes a rhetorical position with no exposed surface. Confront the serious reading and you are told it was a joke. Dismiss the joke and the dread it names is still sitting there, felt by everyone in the room. Irony is how a proposition gets installed without ever being defended, and the flimsy version of this idea has travelled further than any careful one.

## How priming closes the window

Critical futures studies has spent seventy years on the mechanism by which an expectation helps bring about the state of affairs it predicts. The vocabulary turns an intuition into a working instrument.

Fred Polak made the founding claim in *The Image of the Future*: “The image of the future is the key to all choice-oriented behavior.”[^23] Societies are organized by their pictures of what is coming. Where the picture is rich and positive, people build; where it collapses, so does the building. Polak was writing about civilizations, and the same logic runs at the scale of a job decision or a policy debate.

Robert Merton supplied the micro-mechanism in 1948 in his account of the self-fulfilling prophecy: “a false definition of the situation evoking a new behavior which makes the originally false conception come true.”[^24] His own application was to stereotypes, where exclusion produces the disadvantages that the stereotype then cites as proof. Merton's stereotype version is the relevant one here. Believe you are running out of time and you act like someone who has to get through, and enough people acting that way is itself a mechanism of closure. ==Merton's mechanism runs on belief alone, as long as the believers are in a position to act on it.==

Say that concretely, because the mechanism is easy to leave abstract. A graduate chooses proximity to an AI lab because it feels safer than organizing for a broad settlement. An investor moves into the assets expected to capture automation gains. A policymaker designs compensation for an inevitable shock and leaves ownership untouched. An employer treats workforce replacement as the competitive move everybody else will make soon anyway. Each decision is reasonable on its own terms, and together they concentrate ownership, weaken labor power and narrow mobility. ==A future image becomes infrastructure when enough people coordinate their present around it.==

Jens Beckert put the economic version most usefully with fictional expectations, where power is the capacity to create and shape the imagined futures on which others act. Sheila Jasanoff's [[Sociotechnical Imaginaries]] describes how such visions get institutionally stabilized and publicly performed until they define which futures count as legitimate and achievable.[^25][^26] My own [[Condensation of Future Imaginaries]] tracks the process by which a contested field of possible futures thickens into a single one that no longer looks like a choice, and I have used “the AI race” as the standard example.[^27] The permanent underclass is the same structure running on the other emotional register. The race supplies the imperative and the underclass supplies the punishment for losing it.

I should be honest about one thing: nobody has built this bridge in the literature yet. There is decent work connecting Jasanoff's imaginaries to AI anxiety and none that I have found connecting Polak or Merton to the underclass discourse specifically. The connection I am drawing here is mine, and you should treat it as an argument to test.

## The countermove

Sun also wrote the sentence that dismantles the phrase she did most to spread: “The production of a social underclass is a policy choice.”[^3]

That is the entire counter-move. It is a modest one. It leaves both the capabilities of the technology and the fear intact and relocates the question. A permanent underclass is something a society produces, through decisions about ownership, taxation, insurance, education, labor law and who is entitled to a share of what machines make. Every one of those decisions is being made right now, by identifiable people, in rooms you could name. Denaturalizing the claim means putting them back in the frame, which is what [[No future is neutral]] asks of any image of the future: Who wins, who loses, and who is missing from it?

This is also where I would push back on my own field's habit of treating narrative critique as sufficient. Showing that a story serves interests does not make the underlying possibility go away. Myrdal was describing something real in 1963. Wright was describing something real in 1984. Populations do get made economically unnecessary, and what happened to the neighborhoods Wilson described after the factories left is not encouraging. The formation of meme names is available. And that is exactly why the narrative version is dangerous: it takes a genuine political danger and converts it into a personal deadline, so that the people best positioned to prevent it spend their remaining time optimizing their own escape. [[Exit strategies as collective surrender]] is the name for that trade. Even for the people making it, it is a bad one, as Karp's remark about the country blowing up concedes.

So watch the decisions directly. A handful of them are visible from outside the room. Watch whether firms automate tasks or whole roles and whether that shows up in job postings before it shows up in press releases. Watch where the productivity gains land, in wages or in buybacks, and whether any new social insurance stays tied to employment. Watch entry-level hiring in the 22-to-25 cohort, where the one real signal currently sits. Those are the places where a narrated formation would turn into a measured one, and they are also where a narrative critique runs out and ordinary empirical attention takes over.

So the useful reflex, whenever the phrase turns up, is the question in the epigraph and then the follow-up, the one the joke is built to prevent you from reaching: If this outcome required decisions, which ones would be made by whom, and what would it take to make them differently? You can notice when a joke has begun issuing instructions.

Two years is a rhetorical device with a countdown attached. The window may close. How much of it will close because enough people were persuaded that it already had?

## Related concepts and further reading

The ideological neighborhood matters here, because it is the same milieu. The belief systems bundled under [[TESCREAL]], the political architecture of [[The Network State]], the moral arithmetic of [[Longtermism]] and the extractive self-image traced in [[From the Californian Ideology to the Texan Ideology]] all share one operation: classifying human beings by usefulness to a project. A vocabulary that sorts people into the productive and the surplus is already the house language there, which is part of why the phrase settled so easily.

For the futures-studies frame beneath this primer, start with [[No future is neutral]] on the politics of any image of the future, then [[Images of the Future]] and [[Sociotechnical Imaginaries]] for the mechanisms by which such images get stabilized. [[Condensation of Future Imaginaries]] covers the hardening of a contested field into a single obvious one, and [[Reactance and future narratives]] covers what happens when people notice they are being told a future is settled. On the political response the meme forecloses, see [[Exit strategies as collective surrender]].

For the primary sources on the AI version, read Chayka's column and Sun's essay directly.[^2][^3] For the term's history, Gans and Wacquant are the two indispensable critics, and Myrdal's 1963 original is short.[^5][^8][^4]

[^1]: Anonymous meme account on X: “You have 2 years to create a podcast in order to escape the permanent underclass.” Circulating in AI circles from early 2025; quoted and discussed in Chayka's New Yorker column (see [^2]). No verified permanent link to the original post.

[^2]: Chayka, K. “Will A.I. Trap You in the ‘Permanent Underclass’?” *The New Yorker*, Infinite Scroll column, October 8, 2025. https://www.newyorker.com/culture/infinite-scroll/will-ai-trap-you-in-the-permanent-underclass. Source of “A sense of dread about this impending A.I. caste system has created a new urgency to get ahead while you still can” and of the “kind of a joke, but also kind of serious” framing, as well as the lumpenproletariat analogy.

[^3]: Sun, J. “Silicon Valley Is Bracing for a Permanent Underclass.” *The New York Times*, opinion, April 30, 2026. https://www.nytimes.com/2026/04/30/opinion/ai-labor-work-force-silicon-valley.html. Based on more than fifty interviews. Source of “Most people I know in the A.I. industry think the median person is screwed, and they have no idea what to do about it,” of “the production of a social underclass is a policy choice,” of “They muse about the postwork future while pulling 80-hour workweeks,” and of the quoted young tech worker: “Oh, the reason I got to take this job is cuz I got to escape the permanent underclass.”

[^4]: Myrdal, G. (1963). *Challenge to Affluence*. Pantheon. Introduces “underclass” into American debate from Scandinavian usage; defines a growing “class of unemployed, unemployables, and underemployed” increasingly “set apart from the nation at large,” explained through automation and productivity growth rather than behavior.

[^5]: Gans, H. J. (1995). *The War Against the Poor: The Underclass and Antipoverty Policy*. Basic Books. https://eric.ed.gov/?id=ED389791. Traces the 1973 reappearance of the term in *The Public Interest* as a behaviorally defined and racially coded category, and the late-1970s attachment of “permanent” and “trapped.” The Time cover “The Underclass” ran on August 29, 1977, with the phrase “the unreachables; the American underclass”; author unidentified. Cover: https://content.time.com/time/covers/0,16641,19770829,00.html. Edward Kennedy's 1978 use of “a permanent underclass in our society” is the earliest precisely documented instance of the exact phrase; contemporary citation in the *Journal of Sociology and Social Welfare*, https://scholarworks.wmich.edu/cgi/viewcontent.cgi?article=2216&context=jssw.

[^6]: *U.S. News & World Report* warning of a “permanent underclass,” 1979 (no verified online source). *Chicago Tribune*, “In A Nation Of Riches, A Permanent Underclass,” December 1, 1985, https://www.chicagotribune.com/1985/12/01/in-a-nation-of-riches-a-permanent-underclass-3/. Usage of the phrase rises sharply from 1985 onward.

[^7]: Auletta, K. (1982). *The Underclass*. Random House. Wilson, W. J. (1987). *The Truly Disadvantaged: The Inner City, the Underclass, and Public Policy*. University of Chicago Press, using the term descriptively and structurally against the culture-of-poverty thesis, and defending it in print as late as 1989. A possible later distancing in *When Work Disappears* (1996) is unverified.

[^8]: Wacquant, L. (2022). *The Invention of the “Underclass”: A Study in the Politics of Knowledge*. Polity. https://www.wiley-vch.de/en/areas-interest/humanities-social-sciences/sociology-12so/urban-sociology-12so4/the-invention-of-the-39-underclass-39-978-1-5095-5217-7. Source of “racialized folk devil,” “phantom group” and “cognitive tool of symbolic degradation and domestication.” Gans's alternative term “undercaste” appears in his 1990 and 1995 work (see [^5]).

[^9]: Wright, E. O. (1984). “A General Framework for the Analysis of Class Structure.” *Politics & Society*, 13(4). https://doi.org/10.1177/003232928401300402. Source of the underclass as “economically oppressed but not consistently exploited.” The same argument is restated in his *New Left Review* I/202 essay (1993); the exact wording there is unverified, so the 1984 article is cited.

[^10]: Massenkoff, M., & McCrory, P. (2026). “Labor Market Impacts of AI: A New Measure and Early Evidence.” Anthropic Economic Research, March 5, 2026. https://www.anthropic.com/research/labor-market-impacts. Finds no systematic rise in unemployment in AI-exposed occupations since late 2022; the roughly 14 percent decline in the job-finding rate for 22-to-25-year-olds in exposed occupations is described by the authors as “just barely statistically significant.” Exposure skews toward older, higher-paid, more educated knowledge workers.

[^11]: Department for Science, Innovation and Technology with the AI Security Institute, “Assessment of AI Capabilities and the Impact on the UK Labour Market,” January 28, 2026. https://www.gov.uk/government/publications/assessment-of-ai-capabilities-and-the-impact-on-the-uk-labour-market. Source of “The available evidence does not yet provide clear answers to many of the questions that matter most for policy.”

[^12]: International Labour Organization and NASK, *Generative AI and Jobs: A Refined Global Index of Occupational Exposure*, May 20, 2025. https://www.ilo.org/resource/news/one-four-jobs-risk-being-transformed-genai-new-ilo%E2%80%93nask-global-index-shows. Roughly one in four jobs with some generative-AI exposure; “transformation of jobs is the most likely impact,” with no replacement forecast offered.

[^13]: Findings of broad labor-market stability from the NBER, the Federal Reserve Bank of Atlanta, the Census Bureau and the Yale Budget Lab, as compiled in Andreessen Horowitz's May 2026 argument (see [^18]); the Stanford result of a roughly 16 percent relative decline for 22-to-25-year-olds is the best-supported exception and matches the Anthropic finding.

[^14]: Sacasas, L. M., on the “Borg Complex,” the rhetorical reflex of declaring technological change inevitable and resistance futile; Hicks, M., on inevitability claims as a transfer of power to the builders of the technology declared inevitable.

[^15]: Aschenbrenner, L., on the AGI window closing by 2027; Soares, N., advising against reliance on employment; Das, D. (Menlo Ventures), viral post on compounding wealth ahead of the transition, May 2026.

[^16]: Aschenbrenner's fund is named after *Situational Awareness*, his 2024 AGI manifesto. Reuters, “Situational Awareness‘ portfolio sinks 67% in July on AI stock rout, letter shows,” July 31, 2026, https://www.reuters.com/business/finance/situational-awareness-portfolio-sinks-67-july-ai-stock-rout-letter-shows-2026-07-31/, source of the investor-letter quotations (“We came closer to permanent capital impairment than is acceptable to us” and “We let you down this month”). CNBC, “Leopold Aschenbrenner Situational Awareness fund: $45B fire sale,” July 31, 2026, https://www.cnbc.com/2026/07/31/leopold-aschenbrenner-situational-awareness-fund-fire-sale.html, on leverage of up to 400 percent on AI infrastructure holdings including SK Hynix, CoreWeave and Nebius, short positions in software including Adobe, the forced sale of the equity portfolio to Ken Griffin's Citadel at a discount, and assets falling from roughly 45 billion to roughly 10 billion dollars. Earlier gains in 2026 left the fund up roughly 80 percent for the year.

[^17]: Amodei, D., interview with *Axios*, May 28, 2025: the “white-collar bloodbath” warning, with roughly 50 percent of entry-level office jobs eliminated within one to five years and unemployment at 10 to 20 percent. https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic. Asked by CNN for supporting research, he cited none. Restated in *Bloomberg*, June 2026: “Eventually it gets close to a hundred percent.” Contradicted publicly by Daron Acemoglu and by Yann LeCun (“Dario is wrong”). Note: Amodei is not documented using the phrase “permanent underclass.”

[^18]: George, D. “The ’AI Job Apocalypse‘ Is a Complete Fantasy.” Andreessen Horowitz, May 6, 2026. https://a16z.com/the-ai-job-apocalypse-is-a-complete-fantasy/. Describes the permanent-underclass thesis as “the lump-of-labor fallacy … with updated branding”; *Fortune* reported the firm's own regulatory and portfolio interests in that position.

[^19]: Kwalia, “What Is the Permanent Underclass?”, May 8, 2026. https://kwalia.ai/essays/the-permanent-underclass. Source of the “sales pitch” characterization, of the “attention economy … that profits from the urgency … keeping the panic urgent and easy to sell,” and of “The ’permanent‘ framing is rhetorical: it presumes today's institutions will hold while the technology that destabilises them runs to completion, and both can't be true at once.”

[^20]: *City Journal*, May 6, 2026, arguing the inverse case: that the panic itself, channeled into basic-income policy, could produce the underclass it fears.

[^21]: Karp, A. (Palantir), quoted in Sun (see [^3]): “the country could blow up politically, and none of us are going to make any money when the country blows up.”

[^22]: Sun, J., interviewed by Rory Truex, Civic Forum, “Will AI Create a Permanent Underclass?”, July 3, 2026. https://rorytruex.substack.com/p/will-ai-create-a-permanent-underclass. Source of “the permanent underclass is definitely in a lot of ways a meme and a hyperbole … I don't want to say that this is a very common, sincere belief that the majority of people in Silicon Valley have,” alongside her point that the industry's majority expectation of rising inequality and falling mobility is real.

[^23]: Polak, F. (1961/1973). *The Image of the Future*. Elsevier. Abridged translation by Elise Boulding. https://wiki.p2pfoundation.net/index.php/Image_of_the_Future. Source of “The image of the future is the key to all choice-oriented behavior.”

[^24]: Merton, R. K. (1948). “The Self-Fulfilling Prophecy.” *The Antioch Review*, 8(2), 193-210. https://doi.org/10.2307/4609267. Source of “a false definition of the situation evoking a new behavior which makes the originally false conception come true,” and of the application to stereotypes.

[^25]: Beckert, J. (2016). *Imagined Futures: Fictional Expectations and Capitalist Dynamics*. Harvard University Press. https://www.mpifg.de/859363/2016-03-wz-beckert. Power as the capacity to create and influence the fictional expectations on which others act.

[^26]: Jasanoff, S., & Kim, S.-H. (2015). *Dreamscapes of Modernity: Sociotechnical Imaginaries and the Fabrication of Power*. University of Chicago Press. On the application to AI anxiety, see “AI anxiety?” (*Sage*, 2023) and “AI, Risk, and the Power of Imaginaries” (arXiv).

[^27]: Kleske, J. “The Phantom AI Race: A metaphor that runs without competition.” *Futures Lens*. https://futureslens.johanneskleske.com/p/the-phantom-ai-race.

## Sitemap

- [Notes Index](https://garden.johanneskleske.com/notes-index)
- [Futures Terminology](https://garden.johanneskleske.com/futures-terminology.md)
- [Sitemap](https://garden.johanneskleske.com/sitemap.md)