What AI Is Doing To Jobs Is Worse Than We Thought | VisualPolitik EN

Summary

The video argues that AI is unlikely to eliminate work all at once, but it could rapidly reshape office jobs, hiring, wages, and career paths. The business challenge is less a shortage of jobs than moving workers into new roles—and ensuring they can use AI effectively.

Key business implications and playbooks

• Treat jobs as chains of tasks, not bundles of interchangeable work. If AI can handle most tasks but not the final one—such as taking responsibility, exercising judgment, or interacting with people—a human may still be needed. Automating some tasks can also let employees focus on more complex, valuable work.
• Plan for task bundles and possible discontinuities. Rather than assuming jobs will erode gradually, the video suggests that whole roles could change quickly once AI can perform a remaining critical task. Businesses should reassess roles and workflows, not just automate isolated activities.
• Redesign work around human-AI collaboration. Identify where AI adds speed and where human oversight is essential. Employees need to know both how to use AI and where it makes mistakes; occasional use alone may not improve productivity.
• Prepare for workforce mobility. Organizations and workers may need reskilling, career changes, or relocation as office roles contract and demand grows elsewhere. The video emphasizes that transition capacity—not simply the number of jobs—is the central challenge.
• Build AI capability as a competitive skill. The recommended approach is to learn, experiment, understand the technology’s limits, and incorporate it into work. The video also notes that AI can lower barriers to starting a business, including small companies built around a founder and AI agents.

Scenarios and metrics cited

• AI capability: An evaluation lab measured the longest task a model could complete successfully at least half the time. The best model’s task duration reportedly increased from just over 2 hours in August 2025 to about 16 hours by May. If growth continued—even more slowly—the video says AI could handle tasks worth roughly 500 hours of work by 2029. It cautions that these tests mostly involve programming and do not establish that AI can replace a full-time worker.
• Possible 2030 outcomes: Economists working with Anthropic describe three scenarios, without assigning probabilities:
  • Moderate: AI boosts productivity but has little employment impact.
  • Significant: The U.S. economy is 8% larger than it otherwise would be, while office employment is 4% lower.
  • Extreme: AI performs nearly half of current office work, and almost one in five office workers could be left without a job.

The scenarios look similar through 2027, making that a potential point when their paths begin to diverge.

• Early-career hiring: Stanford’s Digital Economy Lab reportedly found employment among 22–25-year-olds in roles most exposed to AI was 19% lower than among peers in less-exposed jobs. The concern is fewer entry-level hires, since routine early-career tasks are especially automatable.
• Job creation and transitions through 2035: McKinsey Global Institute estimates AI could make about 36 million U.S. jobs unnecessary, while economic growth creates about 40 million. However, roughly 11 million workers may need to change fields—more than 3.5 times the normal rate of job change. Only one in seven displaced workers reportedly has a direct route to a growing occupation at a similar salary.
• Wages and the distribution of gains: In the extreme Anthropic scenario, 2030 office salaries are 11.5% lower than they might otherwise be, while pay for occupations such as electricians, bricklayers, and nurses is 34% higher. The video also says labor’s share of U.S. income—about $60 of every $100 today—could fall as output gains accrue more to owners of capital. It cites lower scenario figures of $56 and, in the most extreme case, $45 per $100.
• AI skills in the labor market: PwC figures cited in the video show jobs requiring AI skills pay 62% more, while vacancies requiring those skills grew 69%, compared with 9% growth across the rest of the market. Funcas is cited as forecasting 1.6 million new jobs connected to AI and related businesses.

Examples and evidence

• Microsoft: The company reportedly offered voluntary buyouts to about 60,000 employees, described as roughly a quarter of its workforce, while planning around $190 billion in spending, mostly on AI. The video presents this as a shift in staffing and investment priorities, not simply cost-cutting at a struggling company.
• Radiologists: Despite Geoffrey Hinton’s 2016 prediction that AI would make radiologists obsolete, U.S. radiologists reportedly earned an average of $571,000 in 2025, up 9% year over year, with thousands of vacancies unfilled. The video attributes this to more imaging—scans reportedly rose 25% from 2018 to 2025—plus legal accountability and duties beyond image interpretation, including consultations, procedures, and patient care.
• Customer-support agents: A study of more than 5,000 agents, published in the Quarterly Journal of Economics, reportedly found AI users resolved 14% more requests per hour on average. For beginners, the gain was 34%, suggesting AI can help newer workers close some experience gaps.
• AI can also hinder work when used poorly: In a Meta test, experienced programmers reportedly took 19% longer on AI-assisted tasks despite believing they were 20% faster. This underscores the need to judge when AI is useful and verify its output.
• Physical infrastructure: The video highlights growing demand for electricians, plumbers, installers, technicians, and builders to construct and maintain data centers. It notes that switching from office work to a skilled trade is not straightforward and may require substantial retraining.

Sources and presenters

The presenter is an unnamed narrator for VisualPolitik EN. Sources and people cited include Microsoft; Sam Altman and OpenAI; IMF Managing Director Kristalina Georgieva; economists from the University of Virginia and Anthropic; an AI evaluation lab; Meta; Geoffrey Hinton; economists from the University of Toronto; Stanford’s Digital Economy Lab; McKinsey Global Institute; Goldman Sachs; Jensen Huang and Nvidia; Carnegie Mellon University; PwC; Funcas; and the Quarterly Journal of Economics study of support agents.