The First AI Underclass Will Be Its Builders
They Will Have Access. They Will Have Tools. They Will Own None of the Power.
The permanent underclass will not announce itself through breadlines.
It will appear first as a strange humiliation among people who are still employed, still credentialed, still close to power, and yet increasingly aware that proximity is not ownership.
This is the hidden panic inside Silicon Valley’s AI boom. The shallow version says: AI will take jobs. The deeper version says: AI may destroy the bargaining power of the people who once believed intelligence itself was their leverage.
For a generation, Silicon Valley offered a simple path to sovereignty: become technical, move close to the future, join the right company, accumulate equity, and ride the next platform shift upward. The social contract was brutal but legible. If you were smart enough, early enough, adaptable enough, you could convert labor into capital.
AI threatens that contract from the inside.
You did everything right, and the future still demoted you.
You optimized yourself for the machine. You learned its language. You moved closer to the frontier. You accepted the culture of permanent adaptation. You became faster, more technical, more flexible, more legible.
And now the frontier is adapting you downward.
The people building the machine are beginning to realize that they may not be its beneficiaries. They may be its first absorbed class.
The Panic Silicon Valley Cannot Say Out Loud
The phrase “permanent underclass” recently entered mainstream Silicon Valley discourse because it named something people were already feeling but could not say cleanly.
In one version, it appears almost absurd: a San Francisco debate over whether even people earning below $500,000 a year might be trapped outside true financial independence in the AI capital zone. VC Deedy Das’s viral framing drew ridicule, but it also exposed a real status anxiety: in the AI boom, income is no longer the same thing as escape. The dividing line is not salary. It is ownership of explosive upside.[1]
That is why the debate resonated beyond its obvious Bay Area excess. A person can be highly paid and still structurally demoted. A person can be close to power and still not own power. A person can use AI every day and still watch the surplus flow elsewhere.
The anxiety is not poverty.
The anxiety is revelation.
Silicon Valley workers are beginning to understand that “being technical” may no longer be enough. Code is no longer a private priesthood. Design is no longer protected by taste. Writing is no longer protected by fluency. Analysis is no longer protected by pattern recognition. Coordination is no longer protected by meetings, dashboards, and managerial ritual.
The machine is moving upstream.
The End of “Learn to Code”
The old promise was: learn to code.
The new reality is: code learned to code.
That sentence is more than a meme. It is the collapse of an initiation ritual.
For years, the answer to insecurity was technical adaptation. If your industry was unstable, learn software. If your job was low-status, learn software. If your country was stagnant, learn software. If you wanted to move closer to the future, learn software.
Coding was not merely a skill. It was a migration path into leverage.
AI does not need to eliminate coding to weaken that path. It only needs to make coding less scarce.
That is the wound. The same skill that once marked entry into the future now risks becoming a supervisory layer above systems that generate, test, refactor, and deploy more of the work themselves.
The engineer does not disappear.
He becomes less central.
The junior developer does not revolt.
He never gets hired.
The first injury is not unemployment. It is watching your apprenticeship disappear behind an interface.
Leverage Collapse, Not Job Loss
The permanent underclass of AI will not be defined primarily by unemployment.
It will be defined by leverage collapse.
This distinction matters. A society can still have jobs while destroying the bargaining power, status, and autonomy those jobs once carried. People may continue to work, but their work may no longer command scarcity. They may produce more than ever while capturing less of the surplus. They may become AI-amplified, but not AI-sovereign.
This is the central misunderstanding in the public debate. The question is not simply whether AI eliminates all software engineers, lawyers, analysts, designers, marketers, recruiters, or support workers. The question is whether those workers retain leverage once cognition becomes infrastructure.
Industrial capitalism still needed labor because execution was embodied. Digital capitalism still needed skilled workers because cognition was scarce. AI capitalism attacks the scarcity of cognition itself.
Once cognition becomes infrastructure, the human worker is no longer the primary site of execution. He becomes a validator, narrator, compliance surface, taste-checker, liability shield, brand layer, or temporary patch around the system.
That is not unemployment.
It is demotion from agent to input.
The machine does not need to fire you to weaken you.
It only needs to make your abilities less scarce.
How the Demotion Happens
The transition will not arrive as a CEO announcing that humans are obsolete.
It will arrive as compression.
A hiring plan quietly reduced from twenty roles to six. A junior analyst position that never opens. A support team converted into a dashboard. A designer turned into an editor of generated options. A recruiter managing fewer searches because the company is hiring fewer humans. An engineer spending less time writing code and more time supervising systems that make his own apprenticeship unnecessary.
The firm calls it productivity.
The labor market calls it fewer openings.
The worker calls it anxiety.
The system calls it progress.
This is how structural demotion usually happens: not as a single dramatic replacement event, but as a thousand small decisions that make human development look inefficient.
Junior tasks become AI workflows. Teams shrink. Senior workers become reviewers of machine output. The first rung disappears. The remaining humans become more productive, but the path into becoming one of those humans narrows.
The company becomes leaner.
The profession becomes less regenerative.
The worker becomes faster.
The worker becomes weaker.
This is the paradox at the center of AI labor. The individual using AI may become more capable in the moment, while the class of workers he belongs to loses bargaining power over time.
Productivity is not leverage if everyone receives the same tool and the surplus goes to the owner of the system.
The Builders Who Will Not Own the Machine
The first psychologically exposed class will not be the poorest workers.
It will be the near-elite.
Not the founders of frontier labs. Not the early OpenAI, Anthropic, Nvidia, or xAI equity holders. Not the people who own compute, distribution, capital, or model access.
The exposed class is the group close enough to the machine to understand what is happening, but not close enough to own the upside.
They are engineers outside the frontier labs. Product managers whose coordination work is becoming theater. Designers watching taste become promptable. Recruiters watching hiring freeze. Junior developers losing the first rung of the ladder. Startup operators with tiny equity stakes. SaaS employees watching AI-native companies do more with fewer people. Knowledge workers who still have income, but declining strategic importance.
They are not excluded from the future.
They are included as instruments.
This is why the mood is so strange. Silicon Valley is not experiencing simple despair. It is experiencing a crisis of rank. The old hierarchy said that technical competence created upward mobility. The new hierarchy says that ownership of the constraint stack determines sovereignty.
The constraint stack is compute, models, data, distribution, capital, and institutional permission.
Everyone else receives tools.
Tools are not sovereignty.
The New Caste System of AI
The AI economy is beginning to reveal a sharper hierarchy:
- Compute owners
- Model owners
- Distribution owners
- Capital allocators
- Frontier researchers
- AI-augmented operators
- Prompt-mediated cognitive labor
- Human legitimacy layers
- Service and care labor
- Managed surplus population
Silicon Valley workers fear sliding from levels five and six toward seven and eight.
They do not fear becoming useless tomorrow. They fear becoming decorative humans around an increasingly autonomous execution layer.
That is the core wound.
You optimized yourself for the future, and the future converted you into training data.
You joined the machine as an operator and woke up as a dataset.
You thought intelligence was your asset. Then intelligence was relocated into the stack.
This is why the underclass discussion does not fit neatly into left or right.
The left sees capital concentration. The libertarian sees state incapacity and bureaucratic sclerosis. The accelerationist sees temporary pain on the road to abundance. The doomer sees civilizational fragility. The founder sees productivity. The worker sees replacement. The investor sees margin expansion. The public sees another elite industry insisting that disruption is inevitable while asking everyone else to absorb the cost.
Jasmine Sun’s widely circulated New York Times essay described the mood directly: many people inside the AI industry believe the median person is in trouble, and they do not know what to do about it.[2]
That sentence matters because it captures the private consensus behind the public optimism. The builders are not naive. They are often more worried than outsiders. They can see the capability curve. They can see the institutional lag. They can see how quickly the labor market’s assumptions can break.
Anthropic CEO Dario Amodei has also warned that AI could wipe out large parts of entry-level white-collar work, with some reports summarizing his concern as the disappearance of up to half of such jobs within five years.[3]
The exact timeline is uncertain.
The direction is not.
The first rung is already trembling.
Managed, Not Needed
“Managed surplus population” does not mean people vanish from society.
It means they remain present but lose productive centrality.
They are kept solvent enough to consume, entertained enough to remain calm, credentialed enough to feel included, and monitored enough to be governed. They are not outside the system. They are the population the system no longer needs to negotiate with.
This is the darker implication beneath the polite language of productivity.
A worker with leverage must be bargained with.
A citizen with leverage must be persuaded.
A class with leverage must be incorporated.
But a population without leverage only has to be managed.
That management may be comfortable. It may include subscriptions, benefits, entertainment, public language of inclusion, periodic transfers, therapeutic culture, and endless interfaces designed to keep people engaged with systems they do not control.
The underclass of the AI age may not look like starvation.
It may look like permanent access without authorship.
Included, but not sovereign.
Connected, but not powerful.
Managed, but not needed.
The Ladder Is Being Automated Away
A civilization does not need to eliminate all jobs to create an underclass.
It only needs to destroy the ladder.
Entry-level work has always been inefficient. Junior employees are slow. They ask basic questions. They require mentorship. They handle small tasks before they are trusted with larger ones. From the perspective of short-term productivity, they are expensive.
But from the perspective of civilization, they are how humans become competent.
AI threatens this developmental layer first. If junior coding, junior design, junior analysis, junior support, junior writing, and junior operations can be compressed or automated, companies may become more efficient while society becomes less regenerative.
The firm saves money.
The profession loses its inheritance mechanism.
This is how underclasses form: not only through poverty, but through blocked initiation. People remain outside the guild because the apprentice tasks have been automated away.
The optimistic case is not stupid. Every major technological transition has created new roles, new industries, and new forms of wealth. The internet destroyed old business models while creating software empires, creator economies, cloud platforms, and new kinds of work. AI may do the same.
But the question is not whether humans remain busy.
The question is whether their busyness restores leverage.
The optimist asks whether humans will still have work.
The harder question is whether work will still make humans powerful.
Silicon Valley built its self-image on meritocratic entry. AI may transform it into a world of closed ownership, elite research enclaves, and tool-using dependents.
Access Is Not Power
The most seductive lie of the AI age is that access equals power.
It does not.
Giving everyone a chatbot does not democratize intelligence if the surplus accrues to the owners of the model, the compute, the distribution channel, and the capital structure. Universal access to an interface can coexist with extreme concentration of sovereignty.
A person using AI to produce more output may still become poorer in relative terms if everyone else can produce the same output and the owner of the system captures the margin.
The worker becomes more productive. The firm becomes leaner. The platform becomes more valuable. The model improves. The labor market becomes more competitive. The individual gains speed but loses scarcity.
This is not empowerment.
It is assisted commoditization.
The underclass is not the person without AI.
The underclass is the person whose AI use increases output without increasing ownership.
That is the trap.
The Joyless Winners of the AI Boom
The old Silicon Valley worker believed he was joining the frontier.
Now he wonders whether he was merely helping build the frontier for someone else.
This is why the mood has turned strangely joyless. Even some winners seem uneasy. Das has described a kind of existential dread in the Bay Area, where even rapid wealth creation has not produced obvious happiness, while those outside the winning AI circles fear insecurity and irrelevance.[4]
That unease is not incidental. It is structural.
If AI compresses the relationship between effort and reward, then both winners and losers suffer a meaning shock. The loser feels obsolete. The winner feels unearned acceleration. The near-elite feels humiliated. The founder feels pursued by the next model. The researcher feels morally implicated. The public feels managed.
A society can survive inequality more easily than it can survive illegibility.
People can tolerate hierarchy when they believe the ladder is real.
They revolt, withdraw, or decay when they suspect the ladder has become theater.
Execution Has Left the Human Body
This is execution outrunning legitimacy.
Execution is moving into models, compute, infrastructure, and automated coordination.
Legitimacy remains trapped in the old language of education, work, productivity, merit, citizenship, and career progression.
The visible institution still says:
study → work → income → dignity → citizenship.
The hidden execution layer increasingly says:
compute → model → automation → capital return → infrastructure power.
The gap between those two sequences is the permanent-underclass problem.
Not mass unemployment alone.
A legitimacy failure.
A society built around labor cannot easily explain what humans are for after labor loses bargaining power. A democracy built around citizens cannot easily explain what citizens are for if the economy no longer needs them as productive participants. A meritocracy built around intelligence cannot easily explain status once intelligence is synthetic, scalable, and privately owned.
This is why Silicon Valley’s anxiety matters. The builders are the first to see the contradiction because they live closest to the execution layer.
They are watching legitimacy lag in real time.
The Question Is Not Jobs. It Is Ownership.
The question is not whether AI should be stopped.
It will not be stopped.
The question is whether the surplus of synthetic intelligence becomes a civilizational asset or a narrow ownership regime.
If AI productivity is captured almost entirely by private model owners, compute owners, distribution platforms, and capital allocators, then the underclass will not be an accident. It will be the default output of the system.
If societies want to avoid this outcome, they will need more than retraining. They will need ways for ordinary people to hold claims on the systems that replace their leverage: ownership, dividends, public infrastructure, labor power, and new forms of status that are not tied to obsolete work.
But that requires admitting the real problem.
Retraining alone is not enough if the thing being retrained for is also being automated.
Access alone is not enough if access does not confer ownership.
Productivity alone is not enough if productivity does not confer dignity.
Abundance alone is not enough if abundance arrives without agency.
The political question of AI is not whether humans will still be busy.
It is whether they will still have a claim on the future.
The First Signal of the AI Underclass
The first signal of the AI underclass will not be mass poverty.
It will be a highly educated worker sitting in San Francisco, New York, London, Bangalore, Taipei, or Berlin, using AI every day, producing more than ever, and realizing that none of this has made him more sovereign.
He will still have tools.
He will still have subscriptions.
He will still have dashboards, agents, copilots, workflows, and interfaces.
But the leverage will have moved elsewhere.
Into the model.
Into the compute.
Into the capital table.
Into the distribution layer.
Into the institutions that license, deploy, and govern the system.
He will not be outside the machine.
He will be inside it, as a managed component.
That is the future Silicon Valley is afraid to name.
The permanent underclass will not begin with people who lack access to AI.
It will begin with people who have access to AI, depend on AI, improve AI, and still own none of the power that AI creates.
References
[1] Business Insider coverage of Deedy Das’s “permanent underclass” / sub-$500k San Francisco debate, May 2026.
[2] Jasmine Sun, “Silicon Valley Is Bracing for a ‘Job Apocalypse,’” The New York Times, May 2026; syndicated excerpts appeared in The Salt Lake Tribune.
[3] Dario Amodei’s 2026 public warnings on AI and entry-level white-collar job displacement.
[4] Business Insider coverage of Silicon Valley’s AI wealth boom and reported mood of existential dread, May 2026.