The AI Class Structure

Abstract

Artificial intelligence is not merely a productivity technology. It is becoming a new layer of economic power, cognitive mediation, and institutional control. While early public debate has focused on automation, job displacement, and productivity gains, the deeper issue is the emergence of a new class structure organized around access to, ownership of, and agency over intelligence systems.

This paper argues that AI inequality will not simply divide society between those who use AI and those who do not. Rather, it is creating three structural positions in the emerging AI economy: Intelligence Owners, who control models, compute, data, platforms, and deployment infrastructure; AI-Amplified Humans, who use AI to expand their productivity, learning, creativity, and decision capacity; and AI-Subjected Humans, who are governed, evaluated, displaced, influenced, or economically shaped by systems they neither understand nor control.

These are not permanent social categories. They are fluid structural positions in the AI economy. Individuals and institutions may move between them. A worker can become AI-amplified. A founder can build powerful products on top of systems they do not own. A consumer can use AI to challenge institutional decisions. But at the system level, AI is beginning to organize power around ownership, amplification, and subjection to intelligence systems.

The central risk is not only income inequality. It is agency inequality: the widening gap between those who shape intelligent systems and those whose lives are shaped by them.

1. Introduction

For much of modern economic history, inequality has been explained through ownership of land, capital, labor, credentials, financial assets, and information. The industrial economy divided society around ownership of physical capital. The digital economy intensified inequality through platform power: those who controlled data, distribution, and network effects captured disproportionate value.

AI introduces a deeper transformation.

Information asymmetry was historically about who knew more. Intelligence asymmetry is about who can reason, predict, decide, optimize, and act at scale.

This means AI does not merely make existing markets more efficient. It changes the underlying architecture of power. A person with access to sophisticated AI can learn faster, produce more, negotiate better, code faster, write better, analyze markets, simulate futures, and coordinate action. An institution with proprietary AI systems can evaluate workers, target consumers, automate decisions, influence behavior, and extract value at scales no human bureaucracy could previously achieve.

The central question is therefore not:

Will AI replace humans?

The deeper question is:

Who will have agency over intelligence, and who will be governed by it?

AI is not simply creating a divide between those who use technology and those who do not. It is creating a new structure of economic power organized around ownership, amplification, and subjection to intelligence systems. These categories are fluid at the individual level, but powerful at the system level. The central question is whether AI becomes a force that broadly distributes agency, or an infrastructure that concentrates it behind invisible systems of prediction, optimization, and control.

2. The Emergence of Three AI Classes

The popular version of AI inequality is often framed as:

1. Humans without AI

2. AI systems

3. Humans with AI

This is provocative, but economically incomplete. AI itself is not a social class. AI is a form of capital, infrastructure, and decision power.

A more precise model is three structural positions.

These are not rigid categories. Individuals and institutions may move between them. But as a structural pattern, they describe the emerging hierarchy of the AI economy.

The term “class” is used here not to suggest fixed identities or permanent social categories, but to describe recurring positions of power within the AI economy. At the individual level, these positions may be fluid. A worker can become AI-amplified. A founder can build on top of systems they do not own. A consumer can use AI to challenge institutions. But at the system level, the distinction between those who own intelligence, those who wield it, and those who are governed by it remains economically and politically significant.

This distinction is important because AI has two opposing tendencies. It can democratize access to cognitive tools. But it can also concentrate agency over the systems that shape work, markets, institutions, and public life.

The debate, therefore, should not be framed as “AI will democratize everything” versus “AI will inevitably create a rigid hierarchy.” The deeper question is whether societies build the institutional, technical, and civic infrastructure necessary to convert AI access into actual agency. This revision strengthens the original paper’s claim that the categories are “not rigid” while also responding to objections that “class” may imply excessive permanence.

3. Class One: Intelligence Owners

The first class consists of those who own or control the core means of intelligence production and deployment.

This includes:

• Foundation model companies

• Cloud and compute providers

• Semiconductor and infrastructure firms

• Data-rich platforms

• Enterprise AI platform providers

• Governments with sovereign AI capacity

• Large corporations with proprietary internal intelligence systems

Their power comes from control over several reinforcing assets:

1. Models — the reasoning engines

2. Compute — the scarce industrial capacity required to train and deploy them

3. Data — the raw material of prediction and personalization

4. Distribution — the channels through which AI reaches users and markets

5. Workflow integration — the embedded use of AI inside institutions

6. Feedback loops — the continuous learning generated by usage

7. Governance control — the ability to define what the system can and cannot do

This class does not merely sell tools. It shapes the epistemic environment in which others operate.

In previous eras, capital owners controlled factories. In the AI era, intelligence owners control much of the cognitive machinery through which decisions are made.

Their advantage compounds because AI systems improve through usage, data, feedback, and integration. The more they are used, the more embedded they become. The more embedded they become, the harder they are to replace.

This creates a new kind of monopoly tendency: not only network effects, but reasoning effects.

Reasoning effects occur when AI systems become more valuable because they are integrated into workflows, exposed to feedback, tuned to institutional contexts, and used repeatedly in decision processes. The more an AI system is embedded in a hospital, bank, law firm, logistics network, school, platform, or government agency, the more it may absorb operational patterns, contextual signals, decision rules, and institutional logic.

However, the claim must be made carefully. Models themselves may commoditize. Open-source systems may catch up to proprietary systems in many domains. Raw model intelligence may become cheaper and more widely available.

But even if some models commoditize, power may remain concentrated around compute, data pipelines, workflow integration, institutional embedding, feedback loops, distribution, and governance. The moat is not merely the model. The moat is the system around the model.

This means Intelligence Owners should not be defined narrowly as those who own frontier models. They are those who control the infrastructure through which intelligence is trained, deployed, governed, distributed, and embedded into institutions.

4. Class Two: AI-Amplified Humans

The second class consists of people and organizations that may not own the core AI infrastructure but know how to use it effectively.

These are the AI-native professionals, founders, analysts, engineers, designers, researchers, educators, physicians, operators, and creators who incorporate AI into their daily work.

Their advantage is not merely access. Many people will have access to AI. The advantage comes from fluency, judgment, workflow design, domain expertise, and the ability to combine human purpose with machine intelligence.

The AI-amplified human can:

• Learn faster

• Write faster

• Code faster

• Analyze more deeply

• Create more variations

• Run simulations

• Make better decisions

• Coordinate complex work

• Translate ideas into artifacts rapidly

This class benefits from what may be called cognitive leverage.

In earlier economies, leverage came from tools, capital, teams, or distribution. In the AI economy, leverage increasingly comes from the ability to orchestrate intelligent systems.

This creates enormous opportunity. A single AI-amplified individual can now perform work that previously required a small team. A small AI-native company can challenge incumbents. A founder can prototype, test, publish, analyze, and sell at unprecedented speed.

This is a critical refinement to the three-class thesis. The power of this group may be greater than many inequality narratives assume. The means of intelligence production may remain concentrated, but the means of intelligence application are becoming radically decentralized.

The real economy does not run on models alone. It runs on workflow design, domain expertise, customer understanding, trust, judgment, implementation, integration, distribution, and problem selection. A small team using AI well may outperform a large organization using AI poorly. A founder with strong domain insight may use commodity models to build a highly valuable product. A professional with deep judgment may become far more effective than an institution with expensive tools but weak integration.

This is why the AI economy may not become a simple top-down hierarchy. It may become a contested ecosystem. The Intelligence Owners may control core infrastructure, but they still need AI-Amplified Humans to apply it, adapt it, and generate real-world value. This creates a structural tension:

AI centralizes infrastructure power. AI decentralizes application power.

The future will be shaped by which force dominates.

But this also creates risk. If only a small portion of society becomes truly AI-amplified, the productivity gap between the amplified and the unamplified will widen dramatically.

5. Class Three: AI-Subjected Humans

The third class is the most important and most vulnerable. These are not merely people who “do not use AI. They are people whose lives are increasingly shaped by AI systems without meaningful agency over those systems. They may be:

• Workers evaluated by algorithmic management systems

• Applicants screened by automated hiring tools

• Consumers targeted by predictive personalization

• Patients processed through automated triage systems

• Borrowers scored by opaque credit systems

• Students guided by adaptive learning platforms

• Gig workers managed by platform algorithms

• Citizens profiled by state or corporate surveillance systems

The key feature of this class is not lack of technology. It is lack of power. They live under systems that classify, rank, recommend, price, approve, deny, nudge, and discipline. They may interact with AI constantly, but as subjects rather than authors.

This is the most dangerous form of AI inequality: not exclusion from AI, but inclusion under asymmetric terms. A person may have access to consumer AI tools and still be subject to automated systems in employment, healthcare, credit, insurance, housing, education, public services, and digital media. Their life chances may be increasingly shaped by systems they cannot inspect, challenge, correct, or govern. That is the central vulnerability of the AI-Subjected Human.

6. Access Is Democratizing; Agency Is Not

One of the most important distinctions in the AI economy is the difference between access and agency.

AI access is becoming more democratic. Millions of people can now use powerful AI tools for writing, coding, research, education, design, and analysis. This is real and should not be dismissed.

But access to a consumer AI tool is not the same as agency over institutional AI systems.

A person may have an AI assistant on their phone and still be unable to challenge:

• an automated hiring rejection

• a denied insurance claim

• a credit decision

• a platform ranking

• a dynamic wage algorithm

• a medical triage recommendation

• an algorithmic feed shaping public perception

This is the paradox of the AI era:

Everyone may get a chatbot. Very few control the systems that decide who gets hired, funded, insured, promoted, treated, priced, or trusted. This distinction helps explain why the democratization argument and the inequality argument can both be true. AI can genuinely raise the baseline of individual capability while simultaneously deepening institutional asymmetry.

The question is not whether people can use AI. Increasingly, they can. The question is whether people can understand, contest, redirect, and benefit from the AI systems that shape consequential outcomes in their lives.

7. The Shift from Labor Inequality to Agency Inequality

Each era of capitalism has divided society along a different line.

This changes the nature of inequality. Income inequality is only one symptom. The deeper issue is agency inequality. Agency inequality means unequal ability to:

• Understand systems

• Challenge decisions

• Access knowledge

• Improve one’s condition

• Participate in markets

• Shape institutions

• Make informed choices

• Govern the tools that govern one’s life

A society can survive some income inequality. It cannot easily survive extreme agency inequality. When people feel that invisible systems control their lives, trust collapses.

This is why AI inequality must be treated not merely as an economic issue, but as a civic, institutional, and democratic issue.

8. Why AI Inequality May Be Worse Than Previous Inequality

AI inequality may be more dangerous than previous forms for five reasons.

9. Counter-Agency: AI as a Tool to Challenge AI Power

The same technology that enables intelligence asymmetry can also enable resistance.

Workers, applicants, consumers, citizens, researchers, and activists may use AI to:

• audit opaque systems

• compare outcomes

• detect bias

• draft appeals

• coordinate collective action

• reverse-engineer platform incentives

• analyze contracts

• challenge bureaucracy

• improve bargaining power

• expose misconduct

This matters.

AI does not only empower institutions. It can also empower counter-institutions.

A job applicant can use AI to improve applications. A gig worker can analyze payout patterns. A patient can better understand an insurance denial. A citizen group can coordinate legal challenges. A researcher can detect systemic bias across thousands of cases. A worker collective can use AI to compare workplace practices, organize evidence, and contest unfair treatment.

This is a powerful counterpoint to a deterministic inequality thesis.

But counter-agency is not automatic. It requires:

• access to relevant data

• legal rights to challenge decisions

• transparency into institutional systems

• collective coordination

• public-interest tools

• AI literacy

• contestability pathways

A consumer AI assistant may help someone understand a decision. But if there is no right to explanation, no appeal process, no data access, and no institutional accountability, individual empowerment remains limited. The goal, then, is not merely AI access. The goal is AI-enabled counter-agency. The future of AI inequality will depend in part on whether workers, citizens, consumers, and communities can use intelligence tools to challenge the systems that govern them.

10. Policy Implications: Pro-Agency Governance, Not Compliance Moats

If AI inequality is fundamentally agency inequality, then solutions must go beyond redistribution of income.

They must include redistribution of intelligence capacity, contestability, and institutional agency.

But governance must be designed carefully. Too little governance allows opaque systems to dominate high-stakes domains. Too much poorly designed regulation can entrench incumbents by creating compliance burdens only the largest firms can afford.

The goal is not heavy-handed bureaucracy. The goal is pro-agency governance: making high-consequence AI contestable, auditable, accountable, and compatible with human review. This means governance should not simply ask:

Is this AI system compliant?

It should ask:

Does this system increase or reduce the ability of affected humans to understand, challenge, redirect, and benefit from AI-mediated decisions?

Possible interventions include the following.

11. Implications for Work

The workplace is where the three-class structure may become most visible. Early AI experiments suggested that AI could reduce performance gaps by helping less experienced workers perform better. In controlled settings, this is plausible and often true. But in real-world adoption, higher-income, more educated, and more experienced workers often adopt AI faster.

They have more autonomy, better workflows, greater permission to experiment, and stronger cognitive scaffolding. This means AI may initially widen workplace inequality. Three patterns may emerge:

1. AI-complemented professionals become more productive and valuable

2. AI-managed workers face greater surveillance and optimization

3. AI-displaced workers struggle to transition without meaningful support

The most important workplace divide will be between those who think with AI and those whose work is directed by AI.

Even among AI-amplified workers, there will be a further split:

• those who use AI to extend judgment

• those who outsource judgment to AI

AI raises the premium on judgment. It does not eliminate it.

This is why workforce policy cannot focus only on reskilling. It must focus on helping people cross into the AI-amplified category with real agency, not simply training them to operate inside AI-managed workflows.

12. Implications for Markets

AI will reshape markets by reducing some asymmetries and creating others.

The market impact of AI is therefore double-edged. AI can make markets more transparent or more manipulative. The difference depends on architecture, governance, and incentives. This is why AI markets should not be evaluated only by productivity gains. They should also be evaluated by whether they reduce or deepen agency asymmetry.

13. Implications for Democracy

Democracy depends on citizens having enough agency to understand, deliberate, organize, and contest power. AI challenges this in two ways. First, it can distort the information environment through synthetic media, personalized persuasion, recommender systems, and automated narrative shaping. Second, it can move consequential decisions into invisible technical systems beyond ordinary democratic scrutiny.

If citizens cannot see how decisions are made, cannot challenge automated outcomes, and cannot understand the systems shaping their choices, democratic agency weakens. A democracy cannot be healthy if its citizens are permanently downstream of opaque intelligence systems. The future of democracy therefore depends not only on voting rights or speech rights, but also on cognitive rights:

• the right to know

• the right to contest

• the right to understand

• the right to human review

• the right to participate in systems that shape public life

Democracy in the AI age will require more than access to information. It will require agency over intelligence infrastructure.

14. The Core Choice

AI will not automatically democratize intelligence. Nor will it automatically destroy human agency. Its social impact will depend on the institutional architecture built around it. The central choice is this:

Will AI become a tool that expands human agency broadly, or an infrastructure that concentrates agency narrowly?

If we fail, society may divide into three structural classes:

1. Those who own intelligence

2. Those who wield intelligence

3. Those who are governed by intelligence

If we succeed, AI can become the greatest agency-distributing infrastructure in history.

The future depends not on the technology alone, but on who controls it, who understands it, who benefits from it, who can challenge it, and who has the ability to use it as a tool of self-determination.

The debate around this paper makes one point especially clear: the issue is not whether AI is inherently democratic or inherently hierarchical. It is both. AI democratizes cognitive tools while concentrating structural power. It empowers individuals while enabling institutions to govern them invisibly. It lowers barriers to creation while raising the stakes of ownership, governance, and contestability.

That is why the defining question is not whether machines can think.

It is:

Who holds the structural reins when they do?

15. Conclusion

The deepest inequality of the AI era may not be visible in income statistics alone.

It may appear in subtler questions:

• Who understands the system?

• Who can challenge the system?

• Who benefits from the system?

• Who is optimized by the system?

• Who owns the data?

• Who governs the agents?

• Who can redirect the machine?

• Who is merely processed by it?

The walls of this new structure may not look like castles, factories, or skyscrapers. They may be APIs, models, datasets, ranking systems, decision engines, feedback loops, and automated workflows. But they are walls nonetheless. Every time a person applies for a job, requests a loan, seeks medical approval, works under platform management, consumes an algorithmic feed, or interacts with an automated institution, they are entering this architecture.

AI can become the greatest agency-distributing technology in history. Or it can become the invisible infrastructure of a new inequality. The choice is still open. But the architecture is already being built.

About the Author

Surendra Reddy is an AI infrastructure builder and founder of 451 Ventures. He has held senior roles at Yahoo, Xerox PARC and AT&T Bell Labs, where he worked on large-scale distributed systems and data platforms. His work focuses on AI-native architectures for decision-making, trust and governance in complex systems, spanning finance, food, health and climate.