Over the past few years, AI has become impossible to ignore. But most of the discussion still happens at a fairly shallow level: which jobs will be replaced, which industries will be disrupted, and who needs to learn the latest tools before they fall behind. Those questions matter, obviously. Humans do enjoy staring at the smoke while ignoring the building’s structural damage.
What they often miss is the deeper point: AI is not just changing the division of labor. It is changing the institutional architecture that underpins modern society. That is why I increasingly find it useful to think about this moment through the lens of technological anthropology.
By technological anthropology, I do not mean a simple history of inventions, nor a neutral analysis of technology as a tool. I mean putting technology back into the broader question of how human beings, as a species, organize civilization. From that perspective, the central question is no longer merely whether this or that job will disappear. It becomes something larger: why did a certain civilizational structure emerge in the first place, and is it still necessary now?
That is the question I keep coming back to: in the post-AI era, why might the white-collar class itself begin to lose its necessity?
This sounds provocative, but it does not mean that everyone in an office is about to be fired. Nor is it just another version of the tired “middle-class anxiety” story. The real point is more structural. The institutional foundations that made modern white-collar work necessary may now be rewritten by technology.
In other words, this is not just about wages, layoffs, or labor-market volatility. It is about a deeper transformation in the middle layer of civilization itself. White-collar labor, as a historical form, may be entering a period of evolutionary decline. And in its place, a new role may be emerging: the black collar.
Before going further, it is worth separating this argument from two ideas that are often confused with it.
The first is the middle-income trap, a concept in development economics. That debate concerns countries, not classes. It asks whether a country that has moved from low-income to middle-income status can continue upgrading its industry, technology, and institutions sufficiently to become a high-income economy.
The second is middle-class shrinkage, a familiar theme in sociology and political economy. That story is about the stagnation or erosion of middle-class incomes in advanced societies, often linked to globalization, automation, rising returns to capital, and shifts in public policy. It concerns wealth distribution, social mobility, and the polarization of the social structure.
This essay is not about either of those.
What I want to describe is a different and more fundamental shift: when AI begins to systematically absorb large categories of white-collar labor, especially labor tied to institutional maintenance, complexity management, and symbolic manipulation, then the occupational base of the modern white-collar class itself may contract or be rewritten.
That means the issue is not simply whether societies continue to get richer, nor only whether income falls. It is whether the institutional roles that historically sustained white-collar existence are still necessary at all.
If “middle-class shrinkage” means fewer middle-class people, then what I am talking about is more radical: the civilizational organ on which white-collar life depends may be getting replaced.
We tend to treat white-collar work as a natural feature of any advanced society, as if mature civilizations simply grow large office-based classes the way trees grow leaves. Historically, though, that is not true. White-collar labor is not timeless, and it is not universal. It is a highly specific product of a particular kind of civilization.
White-collar workers matter not because they sit in offices, but because they perform a crucial intermediate function in modern life.
Economically, the white-collar class has long helped sustain large-scale consumer society. Industrial civilization required more than elite consumption at the top and survival at the bottom. It needed a broad, stable, predictable consuming public. Housing, education, healthcare, insurance, automobiles, travel, media, culture, all of these became central modern industries in part because there was a white-collar class able to purchase them. Still, this essay is less concerned with white-collar workers as consumers than with their role as workers.
Politically, the white-collar class has often served as a buffer of institutional legitimacy. The capitalist elite is small. Lower-income groups, facing more immediate pressure, are more likely to make radical political demands. The broad middle, especially the white-collar middle, has often carried the habits that make modern systems stable: tax compliance, procedural patience, investment in legality, acceptance of rules, and belief in gradualism.
But even that is not quite precise enough. White-collar workers do not uphold institutions only because they ideologically believe in them. More often, they uphold them because they are professionally embedded in them. A lawyer invokes the law in court. An accountant certifies compliance in a financial statement. A middle manager implements process and policy. Their labor reproduces institutional legitimacy in practice. This is not merely political support. It is ontological participation.
Organizationally, white-collar workers are the intermediate gears of modern complexity. Lawyers, teachers, engineers, analysts, project managers, HR professionals, civil servants, consultants, auditors, compliance officers, product managers, middle managers, none of these roles directly produce grain, steel, or electricity. But they keep complex systems from collapsing under their own administrative weight. They interpret rules, process information, coordinate workflows, compress uncertainty, and keep institutions running.
So the white-collar class is not just an income category. It is a civilizational role. Economically, it supports consumption. Politically, it cushions legitimacy. Organizationally, it manages complexity.
That is why the issue is not simply whether some office jobs disappear. If the structural basis of this middle layer is altered, then one of civilization’s stabilizing mechanisms begins to loosen.
To explain this clearly, it helps to zoom all the way out.
If I wanted to explain in one simple image why white-collar work is not a natural feature of all social life, I would ask: why are there no white-collar wolves?
Wolves are not stupid. They cooperate. They have hierarchy, strategy, internal order, and forms of leadership. In a loose sense, they even have politics. But everything wolves do remains directly tied to survival: tracking, hunting, guarding, migrating, reproducing, raising young. Even when there is division of labor, it is division within action, not within abstract institutions.
No wolf is responsible for documenting rules, interpreting procedure, managing data, designing organizations, producing reports, assessing legal risk, or coordinating cross-pack systems. The whole structure of wolf life is immediate, embodied, situational, and embedded in the natural world.
Older wolves may guide the pack. But that guidance remains embedded in action. It is not abstracted into codified procedures, externalized knowledge systems, or institutional roles that require dedicated specialists. Their knowledge is never detached from the body and sedimented into documents, bureaucracies, or abstract systems.
That is what I mean by an all-blue-collar structure. Not blue-collar as a status marker, but as a form of labor directly engaged with the physical world.
Wolves do not lack white-collar work because they lack intelligence. They lack it because their social complexity never reached the level where a separate class for processing abstract complexity became necessary.
That matters. It means white-collar work is not a default feature of all societies. It is a product of civilization at a specific threshold of complexity.
To make this clearer, I find it useful to think in terms of a three-layer model of reality.
This model is loosely inspired by Karl Popper’s “three worlds” and by Niklas Luhmann’s systems theory, but I use it here in a more anthropological sense, focused on the evolution of labor forms.
The first layer is the natural world.
This is the physical reality all species face: land, weather, food, energy, ecological systems, objects, materials. The core problem here is survival and physical action. Much of premodern labor was concentrated in this layer.
The second layer is social cooperation.
This includes norms, customs, authority, kinship, tradition, oral commitments, role differentiation, and collective coordination. Here the key issue is not just how humans relate to nature, but how humans relate to one another in organizing shared action.
The third layer is the abstract world of symbols and institutions.
This includes law, finance, administration, contracts, corporate structures, education systems, data infrastructures, media, brands, reports, compliance frameworks, forms, and processes. These are not natural objects, but much of modern civilization operates here.
These three layers are cumulative, not separate. The first provides the basis for survival. The second organizes group life. The third manages the complexity of advanced civilization.
And white-collar workers are, above all, the primary maintainers of this third layer.
Lawyers manage legal texts. Accountants manage recorded abstractions of value. Analysts manage quantified representations of markets. Project managers manage task flows abstracted from organizations. HR manages institutionalized selection. Editors manage symbolic narratives. Consultants integrate and package complexity.
These jobs all share one trait: they do not primarily act on nature itself. They act on the symbolic and institutional systems human beings have built.
From this perspective, white-collar workers are not just people with middle incomes or office jobs. They are the class that processes complexity in the third layer of reality.
A common intuition about AI is that machines should replace the simplest physical labor first, and only much later challenge complex cognitive work. Reality has unfolded differently.
Today’s AI, especially large language models, is still far from mastering the physical world. It does not yet possess anything like robust world understanding, mature spatial intelligence, or truly general-purpose physical capability. Robotics remains clumsy in messy real environments. AGI and ASI remain speculative.
And yet AI is already reshaping white-collar work in profound ways.
That seems paradoxical only if we assume AI enters civilization through the first layer of reality. It does not. It enters through the third.
The native environment of the third layer is language, text, spreadsheets, rules, code, diagrams, forms, reports, charts, symbolic systems. And that is exactly the material modern AI has been trained on. The internet, software repositories, academic papers, legal archives, corporate documentation, policy texts, forums, media, all of it is the accumulated, digitized residue of third-layer civilization.
In a sense, modern society spent decades unconsciously preparing the richest possible training set for AI.
So AI did not first learn how to move through the physical world like a human. It first learned how to function in the symbolic world. That is why it became good first at writing, coding, drafting, summarizing, legal analysis, knowledge retrieval, research support, data processing, reporting, and strategy assistance.
Much of white-collar work consists of encoded complexity. It is complicated, yes, but it is complicated in ways that have already been formalized into rules, language, categories, workflows, and representations. That makes it exactly the sort of thing machines can absorb.
So AI hits white-collar labor first not because white-collar labor is trivial, but because it lives in the domain most accessible to current AI.
Put bluntly: white-collar civilization is not vulnerable because it was too sophisticated. It is vulnerable because it spent generations reorganizing itself into the form most digestible to machines.
If the first phase is AI replacing tasks in the third layer, the second phase may be more significant: AI may begin to reveal that the third layer itself contains vast amounts of historically accumulated complexity that were only ever necessary because human beings are opaque, error-prone, and strategically deceptive.
This is where transaction-cost theory becomes useful.
Markets and organizations do not coordinate themselves for free. Searching for information, negotiating contracts, monitoring behavior, resolving disputes, auditing outcomes, and enforcing compliance all cost time and resources. To reduce those costs, societies created institutions: contracts, compliance systems, audit layers, managerial hierarchies, approval chains, reporting structures.
Those institutions are themselves costly. But historically they were worth it because they reduced even greater costs created by human unreliability.
Legal contracts become longer because humans exploit loopholes. Approval systems become more elaborate because humans abuse power. Audits become denser because humans falsify records. Middle management proliferates because human coordination is limited. Reports and meetings multiply because real-time transparency is weak.
A large portion of the third layer, then, is essentially a human-adaptation layer. It exists to compensate for the fact that human beings are hard to monitor, hard to trust, and capable of deception.
If more of that system is executed by AI, and if AI behavior becomes more auditable, traceable, and governable than human behavior, then much of that inherited institutional complexity may no longer be necessary.
Contracts can become shorter. Compliance can become continuous rather than periodic. Approval can become real-time validation. Reporting can become live visibility. Organizations can flatten.
If that happens, then white-collar workers are not merely losing tasks. They are losing the complexity layer that made their roles necessary in the first place.
That is why the white-collar challenge is not just job displacement. It may be a compression of institutional complexity itself.
Once the third layer begins to compress, the relationship between humans and systems changes.
Agricultural civilization was about managing nature. Industrial civilization was about managing machines. Informational civilization has been about managing institutional complexity.
The post-AI era may be something different again: the management of an increasingly autonomous artificial nature.
By artificial nature, I do not mean a few smart devices. I mean a world of interlinked, AI-driven, semi-autonomous systems: power grids, logistics networks, financial clearing systems, urban infrastructure, supply chains, traffic systems, information ecosystems, production scheduling, and perhaps eventually robotic infrastructures of every kind.
Historically, these systems were machines plus lots of human supervision. In the future, they may increasingly become self-regulating environments.
A useful analogy is the dark factory, a factory so automated that the lights can be turned off because no human workers need to be present. Production continues, but people are no longer physically central to the operation.
Something similar may happen in the symbolic world. As AI takes over more institutional maintenance, humans may no longer spend their days filling forms, coordinating approvals, drafting endless internal reports, and tending procedural machinery. But the system does not disappear. It becomes a self-operating artificial environment.
At that point, the human question changes. It is no longer mainly “how do humans operate machines?” or even “how do humans manage machines through institutions?” It becomes: how do humans coexist with a partially autonomous artificial world?
That is a very different civilizational problem.
Humans may not be better than AI at operating artificial nature. In many complex technical systems, AI will likely surpass humans at real-time optimization, resource allocation, anomaly detection, and predictive control.
But that does not make humans irrelevant. It changes the level at which humans matter.
The key managerial task of the post-AI era may no longer be institutional maintenance. It may become boundary design.
Boundary design means asking questions like these:
Who sets system goals?
Who grants authority?
Who retains shutdown rights?
Who decides rollback procedures?
Who bears responsibility when things go wrong?
Which systems may operate autonomously, and to what degree?
Which require human review?
Which must remain isolated from the physical world?
Which must preserve manual takeover capacity?
In the old world, institutions were built largely to manage unreliable humans. In the emerging world, governance will increasingly focus on constraining the capabilities of autonomous systems.
That is where the black collar appears.
Why “black collar”? Because the role resembles a referee dressed in black on a football pitch. The referee does not play the game. The referee does not score the goals. But the referee enforces the rules, settles disputes, and intervenes when the game crosses a boundary.
The black collar is not the day-to-day operator of artificial systems. The black collar is the boundary manager between human interests and autonomous systems.
That role may include:
White collars keep systems running. Black collars make sure systems do not outrun human control.
If white-collar work declines, where do people go?
The answer cannot simply be “learn AI” or “switch careers.” The change is larger than reskilling. It is a shift in civilizational role. If the old middle layer shrinks, society will not stop evolving. It will grow new organs. The black collar is one of them.
This role has at least three faces.
The first face is the boundary designer.
These are the people who define what autonomous systems are allowed to do, where human review remains mandatory, where isolation is required, and where manual intervention must remain possible. Their work is not one-time rulemaking. It is continuous recalibration.
The second face is the gardener of artificial nature.
An engineer deals with a controllable machine. A gardener tends an evolving ecology. Black collars in this mode are less concerned with local efficiency than with long-term system health. They ask when to prune, when to constrain, when to nourish, when to introduce new capabilities, and when to hold them back.
The third face is the representative of human interests.
This comes closest to ethics, but it is more concrete. AI systems optimize for targets like speed, accuracy, engagement, efficiency, or throughput. Those goals do not automatically align with human flourishing. Someone must ask the questions the system will never ask on its own: Does this optimization erode agency? Does this automation make society more brittle? Does this efficiency come at the expense of dignity, fairness, or resilience?
In that sense, the black collar is the standing reminder that a system exists to serve human beings, not the other way around.
These three dimensions are not separate professions so much as three aspects of one emerging role.
We can now pull the argument together.
Wolf packs do not have white-collar workers because they do not possess the abstract institutional layer that advanced civilization eventually created. Human civilization, through technological development, moved from direct engagement with nature to social coordination and then to a vast third layer of symbols, institutions, and encoded complexity. White-collar workers emerged because that layer required specialists to maintain it.
But AI is the first technology capable not only of participating in that third layer, but of absorbing, optimizing, and potentially compressing it. First it replaces tasks. Then it may simplify the structure that made those tasks necessary.
If that happens, the white-collar class does not merely face job disruption. It faces a deeper loss of structural necessity.
That does not mean civilization ends. It does not mean humans become obsolete. It means a historical phase may be closing.
White-collar work was never the endpoint of civilization. It was one organ in a particular stage of civilizational development. It was necessary for a time. But when the distribution of complexity changes, organs change too.
The question, then, is not simply which jobs survive. It is this:
When the old middle organ of civilization recedes, what new organ grows in its place? Who draws the boundaries? Who safeguards artificial nature? Who ensures that increasingly autonomous systems remain aligned with human life rather than indifferent to it?
That is where the black collar begins.
And that, more than job loss or productivity metrics, may be the real question of the post-AI era: not just how to preserve work, but how to redefine the human role in a world increasingly run by systems we no longer directly operate.
.png){kind=icon}
