The most important thing AI is doing is not replacing people. It is making them legible.
The public conversation about artificial intelligence is still dominated by a familiar drama: software versus labor, automation versus wages, machines versus humans. That framing is incomplete. The more consequential transformation is not that AI will eliminate work in one clean sweep, but that it will render more of human life measurable, forecastable and governable. In that world, the central winners are not the algorithms themselves. They are the institutions that own them.
This is why the most provocative case against AI is not that it will create a nation of unemployed coders, drivers and analysts, though some of that will happen. It is that AI is becoming the operating system of a new kind of power: one that can observe workers continuously, rank them invisibly, nudge consumers silently and discipline citizens at scale. The technology that promises efficiency may instead tighten the grip of surveillance capitalism and digital authoritarianism, whether in the office, the warehouse, the clinic or the state.
That claim sounds abstract until one notices how quickly AI is moving from the realm of assistance into the realm of management. A résumé is screened by software before a human sees it. A warehouse worker’s pace is tracked second by second. A customer service agent is scored on tone, speed and compliance. A driver is routed by systems that know more about demand than the driver does. In each case, the machine does not merely automate a task. It redefines the terms on which the task is performed, and in doing so shifts bargaining power away from the person doing the work.
That is the hidden genius of AI in contemporary capitalism. It is not only a labor-saving device. It is an instrument of asymmetry.
Automation has always displaced work. AI displaces discretion.
Industrial automation replaced muscle. Office software replaced clerical repetition. AI is different because it reaches into judgment, the last refuge of human autonomy in many organizations. Managers have long wanted to standardize the messy, expensive variability of human decision-making. AI gives them a language, and sometimes a tool, to do it.
Consider what happens when a firm deploys predictive systems to decide who gets hired, who gets scheduled, who is promoted and who is flagged as “underperforming.” The result is not simply efficiency. It is the conversion of labor into a dataset that can be optimized from above. Workers are no longer evaluated through a human conversation that allows for context, discretion and appeal. They are compared against a model. The model can be audited, but it is often opaque. It can be corrected, but only by those with access. It can be explained, but rarely by the people most affected.
That asymmetry matters because it changes the psychology of work. A worker who knows they are being watched behaves differently from one who is trusted. A worker who knows their pace, pauses and communications are constantly measured is less likely to experiment, negotiate or resist. AI therefore does not merely substitute for human labor. It reorganizes labor around compliance.
In this sense, automation is not new. The novelty is the granularity of control. Factories once monitored output in batches; digital systems monitor keystrokes, routes, facial expressions and response times. A line of code can now do what a floor supervisor once did, except faster, more consistently and with a memory that never forgets. The employee becomes data exhaust. The boss becomes a dashboard.
Surveillance capitalism is not a side effect. It is the business model.
If AI were being deployed only to perform benign tasks—translating text, detecting fraud, summarizing meetings—the political stakes would be serious but manageable. The larger problem is that many of the most powerful AI systems are built atop an economic logic that rewards extraction. The product is not merely software. The product is human behavior, collected, inferred and sold back as prediction.
That logic has been called surveillance capitalism: a system in which the continuous capture of data becomes a method of accumulation and control. Its defenders like to say users consent, because they click “accept” or enjoy “free” services. But consent in this setting is often theatrical. Most people do not understand the scale of the data collected about them, the inferences drawn from it, or the downstream effects on prices, feeds, opportunities and exposure. The consent is legal; the power is structural.
AI supercharges this model because it lowers the cost of turning raw data into actionable inference. The more behavior is tracked, the more the system learns; the more it learns, the more persuasive it becomes; the more persuasive it becomes, the more data it can collect. It is a feedback loop, and one that tends to favor scale. Once a company has enough users, enough sensors and enough compute, it can build a map of human behavior that smaller rivals cannot easily match. The result is not a competitive marketplace in the old sense. It is concentration.
That concentration has civic consequences. When prediction becomes the main commercial value of the internet, platforms have an incentive to intensify attention, not protect autonomy. They reward outrage because outrage is sticky. They reward compulsive checking because compulsive checking is measurable. They reward emotional volatility because volatility keeps people engaged long enough to be profiled. AI does not create this logic from scratch, but it refines it. It makes manipulation more precise and harder to detect.
“The machine is not merely answering to us. Increasingly, we are answering to the machine.”
The real workplace revolution is managerial, not technological.
Every wave of automation is sold as liberation from drudgery. Yet the labor market often experiences the opposite before any liberation arrives: intensified monitoring, fragmented tasks and weaker worker voice. AI may be following that pattern, but with a crucial difference. Earlier technologies automated processes. AI automates supervision.
That distinction matters because supervision is where power lives. A factory robot can weld a car body. An AI system can decide how many cars a human team should weld, how fast, at what quality threshold and under what disciplinary regime. In logistics, healthcare, finance and education, the temptation is similar: replace the judgment of front-line professionals with model-driven protocols. The justification is always the same—consistency, scale, objectivity. The result is often the same too: less discretion for workers and more leverage for owners.
There is a deeper irony here. AI is celebrated for eliminating human bias, yet in practice it often launders bias through technical procedure. A manager who once made a questionable hiring choice can now say the algorithm recommended it. A company that once enforced punishing quotas can claim the software merely revealed productivity gaps. In this way, AI can make authority appear neutral, even as it hardens inequality.
That is especially dangerous in labor markets already weakened by contracting, gig work and fragmented employment relationships. A worker who is not a full-time employee, who lacks union protection, and who is managed through apps rather than supervisors has little visibility into how decisions are made. AI fits naturally into that environment because it thrives where accountability is weakest.
Digital authoritarianism does not begin with tanks. It begins with data.
The same tools that optimize commerce can also strengthen coercion. States have long sought better ways to monitor dissent, predict unrest and identify vulnerable populations. AI lowers the cost of all three. A government with access to cameras, phone records, location data, online posts and administrative databases can assemble a near-continuous portrait of a society. That portrait can be used for public safety. It can also be used to chill speech, target minorities and preempt opposition.
This is the logic of digital authoritarianism: rule not only by force, but by anticipatory knowledge. The state does not need to arrest everyone. It only needs to make everyone feel visible. When citizens believe their movements, communications and associations are permanently legible, they adapt. They self-censor. They avoid organizing. They learn the boundaries before testing them.
AI intensifies this by making surveillance more scalable and more selective. Facial recognition narrows a crowd to a list of names. Pattern recognition identifies “suspicious” behavior before any offense occurs. Large language models can sort complaints, correspondence and social media at a speed no bureaucracy could match by hand. The old authoritarian state was loud. The new one can be quiet, clinical and automated.
Democracies are not immune. They often import the same technologies under softer language: fraud prevention, border security, public order, risk management. The problem is that systems designed for exceptional cases tend to expand. Once the infrastructure exists, so does the temptation to use it more widely. What begins as a narrow exception can become a permanent architecture of observation.
The ideology of AI is inevitability, and that is precisely why it should be resisted.
Technological change is often described as natural, like weather. That framing is convenient for companies and governments because it turns political choices into technical necessities. But AI’s trajectory is not written in silicon. It is shaped by incentives: venture capital, procurement rules, labor law, privacy enforcement and the distribution of computing power.
The argument that “resistance is futile” deserves particular skepticism because it mistakes deployment for destiny. AI systems are built, trained, tuned and embedded by institutions with specific goals. A model can be optimized for diagnosis, or for ad targeting. It can help a doctor review scans, or help an employer rank applicants. The same technical machinery can be used in radically different moral economies. The choice is not the model’s; it is ours.
That is why the most serious debates about AI are not about whether a chatbot can pass a test. They are about governance. Who audits the systems? Who has access to the data? Who can contest a decision? Who bears liability when the model is wrong? Who owns the infrastructure? Those questions sound bureaucratic, but they are the front line of democracy in a machine-mediated society.
Yet governance is difficult precisely because AI rewards opacity. The systems become more valuable as they become more embedded, and more embedded as they become harder to inspect. Organizations claim trade secrecy. Vendors claim complexity. Regulators claim technical limits. Workers claim they have no recourse. This is how power hides in plain sight: behind interfaces that feel personalized, efficient and frictionless.
The dangerous fantasy is that AI is a neutral tool waiting to be used well. In reality, tools arrive with institutions attached. If the institutions are extractive, AI will be extractive. If they are authoritarian, AI will be authoritarian. If they are monopolistic, AI will centralize power. The technology does not determine the politics. It accelerates the politics already there.
The question is not whether AI will take jobs, but who will own the future it makes possible.
A serious response to AI must begin with an uncomfortable admission: many of the gains will not be evenly shared. Some jobs will disappear. Many more will be degraded. New roles will emerge, but not always where the old ones were, and not always for the same people. That is the standard story of technological change. What makes the current moment different is the combination of automation with surveillance and concentration.
In earlier eras, productivity gains were often accompanied by mass institutions that redistributed some of the benefits: unions, public schools, social insurance, antitrust enforcement, broadcast regulation. Those institutions did not eliminate inequality, but they tempered it. Today, the opposite instinct prevails. Firms ask for more data, fewer constraints and faster deployment. Governments, dazzled by claims of national competitiveness, often comply.
That is a mistake. A society that treats AI only as a race to be won will accept too much private control and too little public oversight. It will confuse faster prediction with better outcomes. It will discover, too late, that a more efficient system can also be a more obedient one.
The proper response is not Luddism. It is institutional seriousness. That means data minimization, not endless collection. It means independent audits, not self-certification. It means meaningful worker protections, not algorithmic management by decree. It means public-interest infrastructure and enforceable privacy rights. Above all, it means refusing the premise that every improvement in prediction is an improvement in freedom.
AI is often described as if it were a mirror held up to humanity. That metaphor is flattering to technologists and misleading to everyone else. A mirror reflects; a system organizes. The emerging AI order is not just showing us who we are. It is deciding, ever more efficiently, who gets to decide for whom. That is not a technical issue. It is the political question of our time.
