While the ability of AI tools to steam engine entire human occupations remains a subject of heated debate, a sobering reality is starting to settle in.

Big tech companies are laying off workers in the thousands, with CEOs expecting the worst and predicting soaring employment rates among college graduates — while gleefully cutting costs at their companies and not looking back.

Consequently, the subject of which jobs will be at the highest risk of being made redundant by AI tech has received intense interest. Most recently, Andrej Karpathy — an OpenAI cofounder, former AI exec at Tesla, and inventor of “vibe coding” — used Bureau of Labor Statistics data and a heavy dose of AI to rate jobs on a scale of zero to ten, where zero is safe from AI, and 10 is most exposed.

After his interactive chart drew plenty of attention, as Fortune notes, Karpathy got cold feet and pulled it down (though an archived version can still be seen here).

“This was a Saturday morning two hour vibe coded project inspired by a book I’m reading,” he tweeted on Sunday. “I thought the code/data might be helpful to others to explore the BLS dataset visually, or color it in different ways or with different prompts or add their own visualizations.”

“It’s been wildly misinterpreted (which I should have anticipated even despite the readme docs), so I took it down,” he added.

“The ‘exposure’ was scored by an LLM based on how digital the job is. This has no bearing on what actually happens to these occupations, which has to do with demand elasticity and a lot more,” Karpathy explained in followup. “People are sensationalizing the visualization tool and putting words in my mouth.”

While we should certainly take his findings with a heavy dose of salt — AI models still suffer from widespread hallucinations and Karpathy himself maintains we should only use vibe coding for rapid iterations and “throwaway weekend projects” — the data tells an all-too-familiar story. Occupations such as construction laborers, janitors, electricians, barbers, and bartenders, may largely be in the clear, whereas accountants, office clerks, customer service reps, and software developers could be the hardest hit.

That’s more or less the conclusion of AI company Anthropic’s own investigation into the matter as well. Earlier this month, the company released its latest findings about the “labor market impacts of AI.” The company’s researchers found that computer programmers, customer service reps, data entry keyers, medical record specialists, and market research analysts were at the highest risk, or “most exposed” to AI.

But whether employment levels are about to be driven off a cliff thanks to the rampant use of generative AI at the workplace remains debatable. As Anthropic points out in its report, “AI is far from reaching its theoretical capability” and “actual coverage remains a fraction of what’s feasible.”

Tech leaders who are conducting mass layoffs and citing AI to justify them have also been accused of trying to distract from corporate bloat and past overhiring, which critics say is the real reason for the job losses.

But if executives are to be believed, the scale of job losses could be staggering. ServiceNow CEO Bill McDermott, for instance, told CNBC last week that he expects unemployment for new college graduates to reach over 30 percent.

In short, as Karpathy’s vibe-coded project hints at, white-collar jobs are facing an existential threat, while more hands-on and often lower-paying occupations could end up surviving the storm — a conclusion that likely won’t be of much consolation to those in the midst of their post-secondary education in software development, accounting, or business administration.

More on AI and employment: Anthropic Announces Jobs Most at Risk From AI