What Is Artificial Intelligence Actually Doing to Jobs? An Honest Look at the Research.
What Is Artificial Intelligence Actually Doing to Jobs? Here Is What the Research Shows — Not the Headlines, the Research.
Coverage of AI and employment tends toward two extremes: AI will eliminate most jobs within a decade, or AI creates more jobs than it destroys. The research supports neither position cleanly. This article looks at what economists and labor researchers have actually found about AI's employment effects — by occupation, by sector, and by education level — and what it means for people planning their careers today.
By NowCastDaily Technology Desk | April 8, 2026 | 12 min read
The question of what AI is doing to employment has become one of the most discussed topics in economics. The problem is that popular discussion of the topic tends to rely on projections and models rather than observed data, which gives it the quality of science fiction rather than economics. This article focuses on what has actually been measured so far — the early, observable effects of AI adoption on employment — and what that evidence suggests about the trajectory.
What the Early Evidence Shows
The most comprehensive analysis of AI's employment effects to date is the MIT Work of the Future task force's ongoing research program. Their 2024 findings, covering the period from GPT-3's public release through the widespread adoption of generative AI tools, found that AI adoption has had measurable effects on employment primarily in three occupational categories: roles involving routine document processing and generation, roles involving image creation and basic graphic design, and roles involving code generation in standardized environments.
In those three categories, the research found reduced demand for entry-level and junior practitioners — the people doing the most routine work within each category. Demand for senior practitioners — people who direct, evaluate, and take responsibility for the outputs — has not declined and in some cases has increased, because AI tools require expert oversight to produce reliable results. The pattern is consistent with previous automation waves: technology compresses the entry-level of knowledge work while raising demand for the senior practitioners who can manage and validate the technology's output.
The Sectors Where Job Growth Is Happening
The most significant job growth associated with AI adoption is in roles that involve deploying, maintaining, and working alongside AI systems rather than competing with them. The Bureau of Labor Statistics Occupational Outlook Handbook 2025-2026 projects the fastest employment growth in: AI and machine learning engineers (31 percent growth projected through 2030), data scientists (36 percent), information security analysts (32 percent), and healthcare roles involving AI-assisted diagnostics (variable by specialty, but broadly positive).
These are not all highly specialized technical roles. The growth in healthcare AI roles, for example, includes medical assistants and clinical data coordinators who work with AI diagnostic tools — positions that require understanding how to work with AI systems but do not require building them. The pattern that emerges from labor market data is that the growing jobs are those involving human judgment in combination with AI capability — not purely human roles, and not purely AI execution.
OpenAI's announcement in early 2026 that its advertising pilot inside ChatGPT had exceeded $100 million in annualized revenue run rate — reported in NowCastDaily's April 2026 coverage — is one visible data point in a broader pattern of AI companies generating revenue at scale, which in turn funds additional hiring in technical, sales, and operations roles. The AI industry itself is a net job creator, even as specific AI capabilities displace specific tasks in other industries.
The Roles Most at Risk — Specifically
Research from the Goldman Sachs economics research team, published in 2023 and updated in 2025, identified specific occupational categories where AI can perform more than 50 percent of current tasks without human assistance. The categories with the highest AI task exposure include: legal research assistants and paralegals (74 percent task exposure), insurance underwriters (68 percent), financial analysts performing routine report generation (61 percent), customer service representatives handling FAQ-type interactions (58 percent), and data entry workers (77 percent).
High task exposure does not mean these jobs will disappear. Goldman's research, and subsequent analysis by Accenture's workforce team in 2025, found that approximately 60 percent of high-exposure roles will be "transformed" rather than eliminated — the human role shifts from performing the exposed tasks to directing and validating AI performance of those tasks. The remaining 40 percent represents genuine displacement risk, concentrated in roles where the task set is narrow enough that the AI-directed version requires minimal human oversight.
The Roles Most Resistant to AI Displacement
Research from Oxford's Future of Humanity Institute and the Brookings Institution identified several occupational categories with low AI displacement risk, based on the specific nature of the tasks involved. The common characteristic of low-displacement roles is the combination of physical presence with social judgment. Healthcare workers who interact directly with patients — nurses, physicians, physical therapists — require physical presence plus real-time social assessment that current AI cannot replicate. Skilled tradespeople — electricians, plumbers, HVAC technicians — require physical dexterity in unstructured environments that robotic systems cannot yet match at the speed and cost of human labor.
Education is also notably resistant. Teachers are not immune to AI tools — AI tutoring platforms are already changing how students learn outside the classroom — but the social, motivational, and developmental functions of in-person teaching are harder to automate than the information-transmission function. The research suggests that teachers who adapt their role toward those non-automatable functions will be more resilient than those who compete with AI on information delivery.
📊 NowCastDaily Analysis
Our analysis suggests the most useful frame for understanding AI's employment effects is not "will AI take my job" but "which parts of my job will AI do, and what does that leave for me to do?" Most jobs involve a mix of tasks — some routine and AI-susceptible, some requiring judgment and social context and resistant to AI. The people who will navigate the AI transition most successfully are those who invest in the non-automatable parts of their role — the parts requiring accountability, interpersonal judgment, physical presence, or original synthesis — while using AI to handle the routine portions more efficiently. The people at highest risk are those whose entire job consists of tasks that AI can now perform at lower cost, with no natural transition to a higher-judgment role adjacent to it. That risk is real and concentrated, but it is narrower than the headlines about AI eliminating most work suggest.
📌 Key Facts
- 36% — Projected employment growth for data scientists through 2030 (BLS Occupational Outlook, 2025-26)
- 77% — AI task exposure for data entry workers — highest of any major occupational category (Goldman Sachs, 2025)
- 60% — Share of high-exposure roles projected to be "transformed" rather than eliminated (Accenture, 2025)
- $100M+ — OpenAI ChatGPT advertising pilot annualized run rate, early 2026
- Low-risk categories — Skilled trades, healthcare direct care, teaching (social/developmental functions)
- High-risk categories — Entry-level legal research, routine financial reporting, data entry, FAQ customer service
⚡ NowCastDaily Bottom Line: AI is not eliminating most jobs. It is compressing entry-level knowledge work while raising demand for senior judgment and for people who can direct, validate, and take accountability for AI outputs. The transition is uneven — concentrated in specific tasks within specific roles rather than entire occupations — and it is moving faster in white-collar information work than in roles requiring physical presence or real-time social judgment. The research does not support either panic or complacency. It supports specific preparation.
Sources: MIT Work of the Future Task Force, 2024 · Bureau of Labor Statistics — Occupational Outlook Handbook 2025-2026 · Goldman Sachs Economics Research — AI and Employment, 2025
NowCastDaily Technology Desk
AI, technology, and the future of work. NowCastDaily.com