AI: The Real Threat Is Office Workers, Not Programmers

What happened?

A Slashdot article highlights a New York Times analysis that challenges the popular narrative: artificial intelligence does not primarily threaten programmers and software engineers, but a broad layer of office workers who often go unnoticed. According to economists at Northwestern University, who recalculated AI exposure metrics considering the total composition of the workforce, the most affected jobs are secretaries, administrative assistants, payroll, and human resources employees. As Michelle Yin, co-author of the study, notes: “The jobs most affected are secretaries and routine office workers, not data scientists or computer scientists.” This finding contrasts with public perception fueled by layoffs at major tech companies like Google, Microsoft, or Meta, where software engineers have indeed been affected, but they represent a minority fraction of total employment. The Northwestern study, still in working paper phase, uses a novel approach: instead of measuring AI exposure based on typical tasks of those already using the technology, it analyzes the complete composition of the U.S. workforce, including sectors where AI penetration is low but automation potential is high.

Why is it important?

This finding reframes the public debate, which until now has focused on high-status professions like lawyers, consultants, or software engineers. However, administrative workers represent tens of millions of jobs worldwide, spread across all sectors and regions. According to the U.S. Bureau of Labor Statistics, there are over 4 million secretaries and administrative assistants, 1.5 million accounting clerks, and 1.2 million HR specialists in the U.S. alone. Many of these positions offer middle-class wages or a pathway to them, similar to what happened with manufacturing jobs before globalization and automation eliminated them. Ignoring this vulnerability could lead to inadequate policies and a lack of preparation for the next wave of labor disruption. Moreover, the geographic concentration of these jobs in non-metropolitan areas makes them especially sensitive: unlike software engineers, who can relocate to tech hubs, office workers have less mobility and fewer retraining options.

Consequences and perspectives

Although economists warn it is too early to know the magnitude of the impact, the speed of AI adoption by companies and the pace of technological improvement make it necessary for policymakers to consider potential effects. A 2023 McKinsey study estimated that up to 30% of hours worked in the U.S. economy could be automated by 2030, with particular impact on administrative and data processing tasks. However, there are also voices reminding that previous technologies created new jobs. An economist at the University of Illinois notes: “It would be prudent not to focus only on what we lose, but also on what we can gain.” AI could increase worker productivity and income, but this requires active retraining and job conversion policies. Historical examples like the introduction of spreadsheets in the 1980s or banking automation in the 1990s show that while routine jobs were eliminated, analysis and supervision roles emerged. However, the current speed of change is unprecedented: ChatGPT reached 100 million users in two months, while electricity took decades to transform industry. Additionally, unlike previous automation, generative AI can perform non-routine cognitive tasks, such as writing reports or analyzing data, broadening the spectrum of vulnerable jobs.

What should readers know?

• Administrative and back-office jobs (customer service, accounting, payroll, HR) are the most exposed to AI automation, according to a Northwestern University study.
• The public debate has overestimated the risk to programmers and underestimated that of office workers, who are a much larger and more vulnerable group. According to BLS data, office and administrative workers total over 20 million jobs in the U.S., compared to about 2 million software engineers.
• Despite layoffs in tech and finance, there is no solid evidence that AI has damaged the labor market as a whole. Economists urge caution against apocalyptic predictions. In fact, the U.S. unemployment rate remains at historic lows (3.7% in 2024), and sectors like hospitality and healthcare continue to demand intensive labor.
• Public policies should focus on retraining these workers and creating new opportunities, not just protecting existing jobs. Programs like those in Denmark or Singapore, which combine continuous training with wage subsidies, could serve as models.
• AI also offers opportunities: it can free workers from repetitive tasks, allowing them to focus on higher-value activities like personalized customer service or strategic decision-making. Companies like IBM are already using AI to automate HR processes, while their employees retrain for data analysis roles.

“The jobs most affected are secretaries and routine office workers, not data scientists or computer scientists.” — Michelle Yin, Northwestern University

In summary, labor disruption from AI will be broader and less elitist than believed. Administrative workers, who have been invisible in the debate, could be the hardest hit. But with appropriate policies, this transition may not be traumatic. As the Illinois economist notes: “History teaches us that technology destroys jobs, but also creates others. The key is to prepare the workforce for new roles.” The time to act is now, before the wave of automation sweeps away millions of office jobs that seem secure today.