Cognitive Work and AI: The Decline of Mid-Level Jobs
Thiago Sebben
10/24/20267 min
The debate over Anthropic’s rumored launch of the Claude Duet personal AI agent, attributed to CBS News, reflects a concern that no longer hinges on any specific product: when AI systems begin carrying out sequences of intellectual tasks, what happens to the jobs that form the first rung of a career ladder? Since no verifiable link to the CBS News report was provided, the launch should not be treated here as confirmed fact. But the question it raises is real—and labor market signals warrant attention.
The AI Index 2026 from Stanford HAI records a decline of more than 4% in employment in occupations highly exposed to AI, including software engineering and customer service. Meanwhile, an analysis of its findings published by Stanford AI Index News points to a roughly 20% decline in employment among software developers aged 22 to 25 since 2024. These figures do not show that AI is immediately eliminating entire professions. They suggest something less dramatic, but potentially more structural: the compression of tasks that once served as an entry point into intellectual work and a way to learn on the job.
Global context and the inflection point in cognitive work
The decisive shift is not simply moving from a professional to a faster tool. It is the ability to delegate interconnected steps in a workflow to AI systems: researching, summarizing, drafting, reviewing, organizing information, and producing a first version of a solution. Autonomous agents expand this capability by carrying out tasks with less ongoing intervention, although they still depend on oversight, clear boundaries, and human validation. For a broader discussion of this transition, see our analysis of autonomous AI agents.
The impact on employment depends less on “total replacement” than on how teams are reorganized. When machines take over routine tasks, an organization can hire fewer people, redistribute responsibilities, or increase output without expanding its workforce. The same shift can mean greater autonomy and productivity for some workers and fewer entry-level opportunities for others.
The erosion of the learning curve
Entry-level tasks—fixing code, preparing reports, organizing data, reviewing documents—are not always just repetitive work. They also help professionals build experience, learn to recognize errors, and develop judgment. If these tasks are automated without companies creating alternative ways to learn, the path into a profession may narrow before any entire occupation disappears.
This risk is not inevitable, but it cannot be resolved with the generic promise that new roles will emerge. The transition requires concrete mechanisms for training and career progression: supervised projects, rotating responsibilities, skills assessments, and opportunities for early-career professionals to learn how to validate AI-generated results. Without these, organizations may save on labor in the short term while weakening the development of the next generation of experts.
Critical Analysis: Geopolitical Forces and Algorithmic Acceleration
The compression of cognitive work is not distributed evenly. It tends to affect tasks that can be described, standardized, and evaluated with relative ease more intensely. Activities involving ambiguity, negotiation, professional accountability, and contextual knowledge are harder to automate completely — although algorithmic assistance can also profoundly change them.
The Stanford HAI AI Index 2026 reports that about one-third of the organizations surveyed expect to reduce staff in the short term as AI automation advances. This is a business expectation, not a picture of layoffs that have already occurred or a certainty for every sector. Even so, this expectation can influence investment, hiring, and team design decisions before automation has produced its full expected impact.
Geographic Asymmetry: Where Value Is Concentrated
AI can be adopted in many regions without its main economic gains being distributed in the same way. Computing infrastructure, access to capital, model ownership, and the concentration of research centers all influence who captures revenue and sets technical standards. According to Lightcast’s analysis of AI workforce trends, California, Texas, and New York account for one-third of AI-related hiring in the United States.
This concentration matters beyond U.S. borders. Countries that adopt systems developed in other markets may gain productivity while remaining dependent on external providers for infrastructure, models, and essential services. The question, then, is not just who uses AI, but who controls the layers that make its use possible — and who can turn productivity into higher wages, industrial capacity, and political influence.
🌍 Geopolitical & Strategic Outlook: AI may spread globally while ownership of the most valuable assets remains concentrated. Without local research, computing, and governance capabilities, adoption expands access to technology but does not necessarily bring sovereignty over its benefits.
Want to discover where your business is losing money on manual tasks?
Our team conducts a free operational AI audit to identify automatable bottlenecks in sales, customer support, and administrative workflows.
Request Free AI Diagnostic at Moove AI →Comparative table of global scenarios and operational impact
The comparison between traditional knowledge work and production supported by AI systems highlights a shift in how tasks and responsibilities are distributed. This is not a fixed boundary: different roles combine automatable execution and human judgment to varying degrees.
| Dimension | Traditional knowledge work | Work supported by AI and agents | Strategic question |
|---|---|---|---|
| Execution | Professionals manually perform repeatable steps | Systems generate first drafts and carry out parts of workflows | Who checks quality and catches errors? |
| Training | Entry-level tasks help build experience | Some of these tasks can be automated | How can we create new learning pathways? |
| Productivity | Limited by individual time and capacity | Can increase for standardized tasks | Who captures the gains, and how are they distributed? |
| Team structure | More professionals can divide up process steps | Smaller teams can oversee more output | Will cost reductions undermine future training? |
| Accountability | Tied to identifiable authors and processes | Can be spread across users, companies, and vendors | Who is accountable for AI-assisted decisions? |
| Economic power | Also linked to labor and local expertise | Relies more on models, data, and infrastructure | How can long-term technological dependency be avoided? |
Stanford HAI also reports productivity gains in standardized tasks, including 26% in code development and 50% in marketing, according to the AI Index 2026. These results do not mean that every team or industry will see the same performance: the effects vary depending on the task, the quality of the tools, and the ability to integrate them into work. But they help explain why companies may reconsider how many professionals they need to deliver certain outcomes.
The question of how gains are distributed arises when productivity increases primarily benefit the owners of models, data, chips, and computing centers, while workers face pressure on wages or opportunities. This shift is neither automatic nor irreversible. It depends on business decisions, labor institutions, competition, regulation, and professionals’ ability to share in the gains generated.
Structural Implications for the Next Decade
The long-term impact will be determined not only by the quality of the models, but also by how companies and societies reorganize learning, accountability, and the distribution of gains. If automation advances faster than institutions can adapt, greater efficiency may coexist with a narrower career pipeline and increased economic concentration.
The talent pipeline and the risk of an experience shortage
If entry-level tasks are taken over by automated systems, a paradox emerges: companies may hire fewer junior employees today and later find themselves with too few experienced professionals to take on complex roles. The decline in employment among young developers is a trend to watch, not proof that this outcome is already inevitable; an analysis by Stanford AI Index News describes a roughly 20% decline since 2024 among software professionals aged 22 to 25.
To prevent the digital economy from weakening its own pipeline of experts, organizations can:
- Identify which automated tasks also served as learning opportunities.
- Create supervised projects that teach junior professionals to evaluate outputs and handle exceptions.
- Define clear responsibilities for those who approve and use AI-generated work.
- Measure not only immediate productivity, but also skills development and career progression.
From executor to orchestrator—without giving up accountability
The value placed on human judgment should not become another cliché. “Orchestrating AI” is neither an abstract skill nor a substitute for domain expertise. It requires knowing how to frame problems, recognize limitations, question outputs, protect data, and take responsibility for decisions. In this sense, algorithmic literacy is an institutional capability: training individuals is not enough if processes do not define when automation is appropriate and who is accountable for its effects.
⚖️ The Ethical & Human Dilemma: If routine tasks disappear before an alternative way to learn by doing emerges, short-term efficiency may deprive workers of experience—and society of professionals capable of judging automated decisions.
FAQ: Frequently Asked Questions
What does the compression of intermediate cognitive work mean?
It means AI systems are beginning to perform portions of operational and analytical tasks previously handled by early-career or mid-level professionals. One possible consequence is that smaller teams can produce more, reducing demand for certain roles without necessarily eliminating entire professions.
Will artificial intelligence cause immediate mass unemployment?
The cited data point to changes in exposed occupations and a decline in employment among certain groups of workers, but they do not support the conclusion that immediate mass unemployment is inevitable. The impact may unfold gradually through fewer hires, redefined roles, and pressure on wages, with significant differences across sectors and regions.
How does cognitive automation affect early-career professionals?
It may reduce opportunities to perform tasks that once served as an entry point and a chance to learn. If companies do not create other ways to provide supervised, hands-on experience, young professionals may find it harder to build the experience needed for roles with greater responsibility.
Why might AI increase the global concentration of wealth?
Because some of the gains may be captured by those who control computing infrastructure, models, and capital. The outcome depends on competition, regulation, intellectual property, and how benefits are distributed: concentration is a structural risk, not an inevitable outcome.
Which human skills are becoming more important?
Domain knowledge, critical judgment, evidence evaluation, handling exceptions, accountability, and the ability to coordinate automated systems. These skills do not replace technical training; they make it more important to know when to trust, when to challenge, and when not to automate.
Leading Transformation with Awareness and Strategy
The rapid evolution of AI, combined with geopolitical competition over infrastructure, data, and talent, is redefining the role of leaders and decision-makers. It is no longer enough to assess efficiency gains: we must understand who controls these systems, how their benefits are distributed, and which human capabilities must not be sacrificed along the way.
Moove AI positions itself as a strategic intelligence partner, helping organizations reflect on ethical governance, organizational transformation, and the thoughtful implementation of AI. Rather than treating automation as inevitable, we need to help organizations make informed decisions, uphold accountability, and build sustainable transitions.
👉 Visit mooveai.com.br to explore Moove AI’s insights and solutions and position your organization at the forefront of responsible innovation.
Navegação
© 2025. All rights reserved by Moove AI.
Empresa
Endereço
Contato
R. José Clementino Bettega, 120 Capão Raso, Curitiba
Moove AI
38.483.416/0001-80