As fears grow that a jobs apocalypse is headed for the economy, Andrei Karpathy has used AI to determine which occupations in the United States are most vulnerable to the technology.
Over the weekend, the co-founder of OpenAI and former director of AI at Tesla posted a chart using data from the Bureau of Labor Statistics that shows how susceptible all occupations are to AI and automation. Different jobs receive scores on a scale of 0 to 10, with 10 being the most at risk.
Although the overall weighted exposure was 4.9, Kalpathy’s data also showed that occupations earning more than $100,000 per year had the worst average score (6.7), while occupations earning less than $35,000 per year had the lowest exposure (3.4).
His graph quickly gained attention online, with many predicting the doom of white-collar work. However, Karpathy quickly deleted the data.
“This was a two-hour Saturday morning atmosphere-coded project inspired by a book I’m reading,” he explained on Sunday Morning’s X. “I thought this code/data might be useful for others to visually explore BLS datasets, color them in different ways and prompts, and add their own visualizations. This was widely misunderstood (I should have expected it, even though there was a Readme document), so I removed it.”
He did not answer questions about how it was misunderstood or what the correct interpretation should be.
Still, the archived version of this chart may not be all that shocking because it reflects what others are saying about how AI will shape the U.S. labor market.
For example, software developers, computer programmers, database administrators, data scientists, mathematicians, financial analysts, paralegals, writers, editors, graphic designers, and market researchers received a score of 9.
That’s because sophisticated AI tools are increasingly being used to crunch numbers and create content, allowing tasks that once took hours, days, or even weeks for knowledge workers to be performed in minutes.
While AI is seen as increasing the productivity of experienced employees, there is growing evidence that companies have less need for entry-level employees. More companies are announcing layoffs and citing AI, but skeptics see it as a scapegoat to correct pandemic-era overemployment.
Meanwhile, construction workers, roofers, painters, janitors, ironworkers, and grounds maintenance workers scored just 1, according to Karpathy’s chart. Similarly, home health aides, nursing assistants, massage therapists, dental hygienists, veterinary assistants, manicurists, barbers, and bartenders received a score of 2.
Earlier this month, AI startup Anthropic published a report titled “AI’s Labor Market Impact: New Metrics and Early Evidence,” which found that actual AI adoption is only a small part of what AI tools can deliver.
Similar to Karpathy’s data, Anthropic’s paper states that AI could theoretically cover most tasks in business, finance, management, computer science, mathematics, legal and office management roles. While the adoption of AI remains slow, Anthropic said the workers most at risk are older, more educated and higher-income workers.
And earlier this year, a viral essay by Citrini Research painted a catastrophic picture of an economy disrupted by AI, causing a stock market crash.
However, Citadel Securities quickly debunked this doomsday scenario in a blistering report, pointing out that Indeed job data shows demand for software engineers is actually increasing by 11% year over year through 2026.
Citadel Securities also said that routine use of generative AI in work remains “unexpectedly stable” and there is currently “little evidence of an imminent risk of displacement.” Instead of the economy collapsing, new business formation is rapidly expanding in the United States, and the construction of large-scale AI data centers is currently causing a localized construction employment boom.
Additionally, if automation grows at the breakneck pace that Citrini fears, the demand for computing will inherently increase, pushing up marginal costs.
“When the marginal cost of computing exceeds the marginal cost of labor for a particular task, no substitution occurs and a natural economic boundary occurs,” Citadel Securities said.
