The concept of an “AI underclass” has become an undeniable economic reality. For decades, it was believed that higher education and specialised office work provided a permanent shield against automation. Machines were considered to primarily replace manual labour. However, the current trajectory of artificial intelligence suggests a different outcome.
Instead of targeting physical tasks, agentic AI focuses on the ability to reason, plan, and execute complex workflows. This shift is creating a deep structural divide in the global workforce. Many professionals who previously felt secure now find their skills being commoditised by software that can replicate their output at a fraction of the cost.
This resource provides a comprehensive analysis of this shift, the factors behind it, and the strategic pathways for individuals and policymakers to discover new possibilities in an era of unprecedented economic bifurcation.
What is the “AI Underclass”?
The AI underclass is a socioeconomic segment created as a result of AI replacing skilled workers’ jobs and leaving them in unstable employment conditions. Rapid advancements in AI and ML have enabled it to perform tasks of various natures and automate workflows, which eliminates the need for corporations to hire a large number of staff for carrying out daily operations.
This has led to the displacement, downgrading, or commodification of jobs and the emergence of a new subclass that we now call the “AI underclass”. It’s more a structural displacement across industries than an economic regression, leading to unemployment.
AI underclass is characterised by a long-term decline in the economic value of certain skills, as technology can now perform them more cheaply, faster, and at a larger scale.
The 75th Percentile Rule
Economic displacement is often driven by the 75th percentile rule. An AI system does not need to outperform the world’s leading experts to be disruptive. It only needs to perform better than 75% of the people currently in a given role. Because software is more adaptable and requires no benefits or downtime, a sufficient algorithm becomes the logical economic choice for most organisations.
Why Experts Warn About a “Permanent Underclass”
The emergence of this underclass is not merely a technological side effect; it is a result of the current distribution of capital. Economists like Daron Acemoglu argue that if AI is used solely to automate existing tasks rather than creating new and high-value human roles, the result is a permanent economic bifurcation.
- The AI Elite: A small group of individuals and corporations who own the data, the hardware, and the primary intelligence models.
- The AI Underclass: A broad population competing for a shrinking pool of roles that have not yet been automated.
The Death of the Entry-Level Role
One of the most significant impacts is the erosion of the junior position. Traditionally, these roles served as the training ground where workers gained the experience necessary to become experts. For instance,
- Software Engineers: Senior developers now use AI to handle the tasks previously assigned to two or three junior assistants, effectively closing the door on new entrants.
- Data Analysts: Automated tools now handle data cleaning and basic reporting, which were once the primary responsibilities of entry-level staff.
- Legal and Administrative Staff: Tasks involving the synthesis of documents or routine drafting are being handled by models that work instantly.
The Technological Transition: From Generative to Agentic AI
To understand why this is happening now, we must look at the transition from Generative AI to Agentic AI.
Generative AI (2023–2024)
These systems were essentially sophisticated “autocomplete” tools. They could write an email or draft a basic report, but they required constant human prompting and oversight. The human remained the pilot responsible for every turn and decision, while the AI acted as a high-speed library assistant.
Because these early models frequently hallucinated or lost context, they necessitated rigorous oversight. This indicates that the efficiency gains were often offset by the time spent fact-checking and refining the output.
This era was defined by “Human-in-the-Loop” systems where AI could suggest, but never independently conclude a professional workflow.
Agentic AI (2025–2026)
Current AI systems, also known as “AI agents”, act as digital employees. They can reason through multi-step projects, break down a high-level goal into a series of sub-tasks, use software tools independently, and correct their own mistakes without human intervention.
These agents are capable of monitoring their own progress and correcting their own mistakes without human assistance. The system has evolved from a reactive chatbot interface to a proactive workflow.
Agentic AI has fundamentally changed the human role from creator to manager. AI agents can now handle entire business processes, such as managing a supply chain or developing a software component from scratch.
In 2026, Anthropic CEO Dario Amodei and other leaders observed that these agents are now capable of absorbing entire workflows. This shift has changed the unit of work from a single task to a full project and made many entry-level and mid-tier roles unnecessary.
How AI Is Restructuring Organisations
The use of artificial intelligence has reorganised work structures across multiple sectors by automating routine cognitive operations that used to be handled by entry-level and mid-level staff. Tasks that once required teams of junior employees can now be executed by AI systems in seconds.
When a firm adopts an AI agent capable of managing 75% of a junior employee’s tasks, that role inherently disappears. The displaced worker is often forced into the service economy or manual labour sectors where AI still lacks the physical dexterity or presence to compete effectively.
The Formation of New Worker Categories
With the growing adoption of AI tools, the organisational structure is being split into new categories of workers: those who control AI systems, those who work alongside them, and those whose roles are gradually declining. This shift is rapidly impacting companies’ hiring processes, candidates’ career paths, and the permanence of many professional roles.
It’s creating a new workforce pattern. Companies require fewer junior employees to perform these tasks, which is beginning to compress traditional career pipelines. On one end are professionals who supervise and integrate AI into business operations. Their productivity increases because AI expands their output.
On the other end are workers whose roles consist primarily of repetitive digital tasks that AI can replicate. For these workers, wages and job opportunities may decline as automation reduces demand.
White-Collar Dismissal
One of the most visible impacts of AI is the displacement of white-collar roles. Tasks such as basic coding, legal research, marketing content creation, report writing, and data analysis can now be completed promptly using AI tools.
Entry-level roles, which were once the starting point for many careers, are particularly vulnerable in today’s era when AI has dominated many roles. This has led to a phenomenon that many analysts describe as a “hollowing out” of the middle workforce.
The AI Tasker Class
At the same time, a new category of workers has surfaced behind the scenes of the AI economy. These workers perform data annotation, content moderation, and model training tasks that help AI systems learn and improve.
These “AI taskers” operate through gig platforms or outsourcing networks and play a crucial role in supporting AI infrastructure, but often earn low wages and experience unstable employment.
Middle-Class Job Compression
The middle layer of the workforce is gradually shrinking. Since AI-based automation can handle more routine cognitive work, many traditional middle-income, executional jobs face pressure.
Financial modelling, customer service, marketing campaigns, and documentation workflows can all be automated with AI, which means fewer employees are needed to complete the work. Consequently, organisations prefer to maintain smaller teams and rely on AI to enhance productivity and reduce operational costs.
Which Jobs Are Most at Risk?
A variety of jobs are affected by AI, depending on the tasks they involve. The highest risk is associated with jobs that require routine digital work, since these tasks follow predictable patterns. AI systems are often faster and more efficient at producing text, reports and analysing data. As a result, companies require fewer workers to accomplish the same tasks.
Entry-level knowledge jobs are most affected by this change. The work of junior professionals is often repetitive, including drafting reports, reviewing documents, or preparing presentations. AI can now complete many of these tasks in seconds that used to take days or even weeks. It is more difficult for freshers to enter professions such as law, finance, marketing, and software development when fewer junior positions are available.
Many industries are already experiencing this pressure. Marketing and media can use AI to generate blog posts, product descriptions, and email campaigns. Legal services use AI tools for document review and case law summarisation. As part of finance and research roles, artificial intelligence analyses data and produces reports that previously required manual effort. The decision-making and strategy process is still supervised by human professionals, but the execution of these strategies requires fewer people.
Task repetition is a common theme across these roles. Automating work that follows structured processes and produces digital outputs is easier. Artificial intelligence is far less likely to replace jobs requiring complex judgment, physical skill, or direct human interaction.
| Risk Level | Job Category | Reason |
| High | Entry-level coding | AI generates functional code quickly |
| High | Content writing | Generative AI produces editable drafts |
| High | Legal research | AI summarises and analyses case law |
| Medium | Marketing managers | Strategy still required, execution automated |
| Medium | Financial analysts | AI assists, but oversight remains necessary |
| Low | Skilled trades | Physical dexterity difficult to automate |
| Low | Therapists | Empathy and trust cannot be automated |
| Low | Surgeons | High-stakes physical complexity |
Jobs Least Likely to Be Automated
Some jobs remain difficult to automate because they require abilities that AI cannot easily replicate. These roles involve complex decision-making, physical coordination, or direct human interaction. In these environments, responsibility and real-world conditions matter as much as technical knowledge.
Work that requires physical skill is one example. Electricians, plumbers, construction workers, and other skilled trades operate in unpredictable environments. Each task may require different tools or safety considerations. AI can assist with planning or diagnostics, but it cannot easily perform the on-site physical work.
Roles that rely on human judgment also face lower risk. Surgeons, engineers, and senior decision-makers must evaluate risks, make critical choices, and take responsibility for outcomes. AI can analyse data and provide recommendations, yet final decisions still require human accountability.
It is also harder to automate jobs that are based on trust and interpersonal relationships. Health care professionals, such as therapists, teachers, and nurses, work directly with patients. As part of their job, they must be able to communicate effectively, empathise, and understand social and emotional contexts. Professionals may benefit from AI, but replacing the human relationship is more challenging.
Economic Consequences
Artificial intelligence changes how value is created and distributed in the economy. A company’s productivity increases when machines take over tasks previously performed by humans, while the demand for certain types of labour decreases. Economic gains are distributed across industries and affect wages, employment patterns, and income distribution.
Wage Pressure in Routine Knowledge Work
Jobs involving repetitive digital tasks are the first to experience wage pressure. AI systems can now complete basic tasks such as writing, reporting, and coding in a fraction of the time. When companies require fewer workers to produce the same output, the demand for those roles declines, and wages often follow.
Concentration of Economic Gains
AI creates value for companies that build or control the technology. Organisations that develop AI models, operate data infrastructure, or supply computing power capture significant economic returns. This can concentrate wealth within a smaller group of technology companies and investors.
Labour Market Polarisation
There may be a widening gap between high-value and low-value work due to AI. In addition to increasing their productivity, professionals who focus on strategy, system design, and decision-making often benefit from AI. However, jobs that require physical labour or human interaction are still needed. Middle-skilled jobs, particularly routine administrative work, may face the greatest threat as automation reduces repetitive tasks.
Changing Career Pathways
Changes in society also affect career development. Many professions have traditionally relied on entry-level positions where workers gained experience through basic tasks. Artificial intelligence may replace more junior positions, as their work is being carried out by machines. Over time, this could change how workers enter industries and develop their professional expertise.
Is This Just Another Tech Panic?
Concerns about job loss often arise during major technological shifts. History shows that technology can eliminate some jobs while creating new ones. The Industrial Revolution reduced demand for many craft workers but expanded employment in manufacturing and engineering positions. The rise of the internet disrupted print media and some retail roles, yet it also created new industries such as digital marketing and e-commerce.
AI follows a similar pattern of disruption, but the pace is faster. Earlier technological changes spread gradually across industries. AI systems can improve quickly and scale through software platforms, allowing automation to affect many sectors in a short span of time.
Another difference is the type of work affected. Previous waves of automation mainly targeted physical labour. AI focuses on cognitive tasks such as writing, coding, research, and data analysis.
At the same time, new roles are emerging as businesses integrate AI into operations. These include professionals who design AI workflows, manage human-AI collaboration, and supervise how automated systems are used. The long-term impact will depend on whether AI is used primarily to replace labour or to enhance human productivity.
How to Avoid Becoming Part of the AI Underclass
Shift from Execution to Strategy
Many roles are built around producing outputs such as reports, content, code, or analysis. These tasks are increasingly automated because AI systems can generate similar outputs quickly and cost-effectively. Professionals who focus only on execution face growing competition from automated tools.
The more resilient position is strategic work. This includes deciding which problems to solve, evaluating the quality of outputs, and determining how information should be used in decision-making. AI can generate options, but it does not bear responsibility for the outcomes. Workers who guide priorities for the interpretation of results and make final decisions remain valuable because organisations still rely on human wisdom and accountability.
Develop AI Leverage
Learning to work with AI tools is becoming an important advantage in many industries. Rather than competing with automated systems, professionals can use them to increase productivity and expand their capabilities. This often involves integrating AI into everyday workflows such as research, data analysis, content production, and project management.
Skills in system design and workflow automation allow individuals to organise multiple tools into efficient processes. When one person can produce the output that previously required several employees, their value increases. AI literacy, therefore, becomes less about using individual tools and more about understanding how to structure tasks and workflows around intelligent systems.
Build Domain Depth
AI performs best when tasks follow clear patterns and rely on general knowledge. Deep expertise within a specific industry is harder to replace because it requires context, experience, and an understanding of complex systems.
Professionals who develop strong domain knowledge can identify problems, evaluate risks, and interpret information in ways that automated systems cannot easily replicate. In-depth knowledge also enables individuals to guide how AI should be applied in their field. This combination of expertise and technological awareness increases long-term resilience. For example, an experienced financial analyst understands market dynamics and regulatory environments that automated data analysis alone cannot provide. So his job will not be easily replaceable.
Stack Complementary Skills
The most competitive professionals continue to combine technical literacy with human-centred capabilities. Skills such as communication, negotiation, leadership, and systems thinking become more important as organisations rely on both people and AI systems to operate effectively.
Communication helps translate complex information into decisions that teams can act on. Negotiation and leadership remain essential in corporate environments where multiple stakeholders are involved. Systems thinking allows individuals to understand how different processes, technologies, and teams interact. When these skills are combined with technical understanding, workers can coordinate human and automated systems more effectively. This combination of capabilities is much harder to automate and replace than a single technical skill.
Own Assets
Another way to reduce exposure to automation risk is to build assets that generate value beyond hourly labour. Assets may include intellectual property, equity in businesses, digital products, or platforms that distribute knowledge or services. These forms of ownership allow individuals to benefit from their value rather than relying entirely on wages.
For example, creating proprietary research, developing software tools, or building an online audience can generate income that is not directly tied to time worked. As automation lowers the cost of producing many services, ownership becomes a more important source of economic security and independence.
In this environment, long-term resilience depends less on mastering individual AI tools and more on building strategic expertise through adaptability and ownership.
I explore how work, ownership, and value are being reshaped over the next decade in my audiobook, The Future of Work.
Conclusion
The origination of an AI underclass represents a fundamental decoupling of productivity from human labour. Professionals are more concerned about the unpredictability of where they stand in this scenario rather than the fear of AI replacing their jobs.
The traditional ladder of career progression is being eroded as agents commoditise entry-level cognitive functions. In order to remain relevant, a task-executor must become an expert professional who can provide the one thing AI cannot: accountability.
Ultimately, the long-term stability of the global economy will depend on our ability to implement policies that redistribute the immense efficiencies of AI and ensure that technological advancement results in broad societal empowerment rather than a permanent state of economic division.
