While the ability of AI tools to “steam engine” entire human occupations remains a subject of heated debate, resonating with historical anxieties reminiscent of the Industrial Revolution, a sobering reality is undeniably starting to settle in across boardrooms and dinner tables alike. The metaphor of the “steam engine” here isn’t merely poetic; it evokes a fundamental shift in the means of production, a technological upheaval that promises to reshape the labor landscape as profoundly as mechanical power once did, displacing old crafts and creating new ones in its wake. The question today is not *if* AI will impact jobs, but *how deeply*, *how quickly*, and *which ones*.

This growing unease is fueled by tangible evidence: Big tech companies are laying off workers in the thousands, a trend that began with post-pandemic recalibrations but has recently been increasingly framed by executives as a strategic pivot towards AI-driven efficiency. This narrative often casts AI as both the reason for the layoffs and the solution for future growth, creating a perplexing dilemma for the workforce. CEOs, far from expressing concern, appear to be expecting the worst for certain demographics, with some like ServiceNow CEO Bill McDermott predicting soaring unemployment rates among college graduates – particularly Gen Z – within the next couple of years. This stark forecast paints a grim picture for those entering the job market, suggesting that the traditional pathways to professional success are rapidly eroding. Meanwhile, these same companies are gleefully cutting costs at their companies, seemingly not looking back at the human capital they are shedding, often citing AI as the primary driver for these “efficiencies.”

Consequently, the subject of which jobs will be at the highest risk of being made redundant by AI technology has received intense, almost feverish, interest from economists, policymakers, and the general public. Everyone, from seasoned professionals to students choosing their majors, wants to know where they stand in this rapidly evolving landscape. Most recently, this curiosity was piqued by Andrej Karpathy – a prominent figure in the AI world, an OpenAI cofounder, former AI executive at Tesla, and the inventor of “vibe coding” – who ventured into this speculative territory. Karpathy utilized publicly available Bureau of Labor Statistics (BLS) data, coupled with a heavy dose of AI analysis, to rate various jobs on a scale of zero to ten. In his framework, zero represented occupations deemed relatively safe from AI-driven disruption, while ten signified those most exposed to potential automation or replacement.

Karpathy’s interactive chart, a visual representation of his AI’s assessment, immediately drew significant attention, rapidly circulating across social media and tech news outlets. As Fortune notes, the project garnered widespread engagement, with individuals anxiously searching for their own professions. However, the unexpected viral nature and the strong reactions it elicited seemingly prompted Karpathy to get cold feet. He subsequently pulled the chart down from public view, illustrating the delicate balance between academic exploration and public interpretation in the age of rapid information dissemination. Despite its removal, an archived version of the chart can still be accessed, providing a glimpse into the data that caused such a stir.

In a series of explanatory posts on social media, Karpathy clarified the origins and intent behind his short-lived project. “This was a Saturday morning two hour vibe coded project inspired by a book I’m reading,” he tweeted on Sunday. He emphasized that his primary goal was not to make definitive predictions but to offer a tool for exploration. “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.” This statement highlights a common tension in AI development: the creation of powerful analytical tools that, when released to a broad audience, can be easily misconstrued outside their intended context or without the necessary technical understanding.

The swift public reaction and subsequent misinterpretations forced his hand. “It’s been wildly misinterpreted (which I should have anticipated even despite the readme docs), so I took it down,” he added. This admission underscores the challenge of communicating complex AI-driven insights to a non-expert audience, especially when the subject matter directly impacts people’s livelihoods and future prospects. The inherent sensationalism of “AI taking jobs” narratives often overshadows the nuanced methodological details and researcher caveats.

Further elaborating on the core mechanism of his analysis, Karpathy explained, “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.” This clarification is crucial. His model did not predict *actual job loss* or *economic displacement* but rather the *digital susceptibility* of a role. A job that is highly “digital” might involve tasks that can be easily automated or assisted by AI, but this doesn’t automatically translate to job redundancy. Factors like human interaction, creativity, physical dexterity, regulatory constraints, and market demand elasticity play equally, if not more, significant roles in determining the ultimate fate of an occupation. “People are sensationalizing the visualization tool and putting words in my mouth,” Karpathy concluded, expressing frustration at the media’s tendency to simplify complex research into alarmist headlines.

While we should certainly take Karpathy’s findings with a heavy dose of salt – AI models still suffer from widespread “hallucinations” or factual inaccuracies, and Karpathy himself maintains we should only use vibe coding for rapid iterations and “throwaway weekend projects” – the raw data, even with its caveats, tells an all-too-familiar story. The pattern suggested by his chart, despite its informal nature, aligns with many other analyses regarding AI’s impact. Occupations that primarily involve physical labor, direct human interaction, or highly specialized, non-repetitive manual tasks appear to be largely in the clear. Roles such as construction laborers, janitors, electricians, barbers, and bartenders, which often require on-site presence, dexterity, problem-solving in dynamic physical environments, or personal service, may largely be insulated from direct AI replacement in the near future. These jobs often possess a high degree of “analog” interaction that current AI struggles to replicate.

Conversely, the analysis hinted that professions heavily reliant on information processing, data analysis, repetitive administrative tasks, and even certain aspects of creative digital work could be the hardest hit. Accountants, office clerks, customer service representatives, and software developers were flagged as highly exposed. These roles often involve structured data, pattern recognition, communication that can be scripted or templated, and code generation – areas where large language models (LLMs) and other AI tools demonstrate rapidly increasing proficiency. The ability of AI to sift through vast datasets, generate reports, handle routine customer inquiries, or even assist in writing and debugging code poses a direct challenge to the traditional scope of these white-collar jobs.

This conclusion is not an isolated one. It’s more or less the same outcome of AI company Anthropic’s own comprehensive investigation into the matter as well. Earlier this month, the company, a leading competitor in the generative AI space, released its latest findings about the “labor market impacts of AI.” Their rigorous research, distinct from Karpathy’s “vibe coded” approach but arriving at similar conclusions, identified several professions as being at the highest risk, or “most exposed,” to AI. These included computer programmers, customer service representatives, data entry keyers, medical record specialists, and market research analysts. The consistency across different methodologies and research entities strengthens the argument that certain job categories are indeed more vulnerable to AI-driven automation and augmentation.

However, whether employment levels are truly about to be driven off a cliff thanks to the rampant use of generative AI at the workplace remains a subject of intense debate among economists and industry experts. While the capabilities of AI are impressive, their practical deployment and integration into workflows are complex and often slower than anticipated. 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.” This suggests that while AI has the *potential* to automate many tasks, the current reality is that it often serves as an *assistant* or *augmentative tool* rather than a complete replacement, at least for now. The gap between what AI *can* do in a lab setting and what it *does* in a real-world corporate environment is significant, tempered by factors like cost, legal implications, ethical considerations, and the sheer inertia of organizational change.

Adding another layer of complexity to the discussion, tech leaders who are conducting mass layoffs and concurrently citing AI to justify them have also been accused of trying to distract from more mundane, yet potent, corporate issues. Critics suggest that AI is being used as a convenient scapegoat for past instances of corporate bloat and overhiring, particularly during the pandemic-driven tech boom. During that period, many companies rapidly expanded their workforce to meet unprecedented demand, only to find themselves overstaffed when market conditions normalized. Attributing layoffs solely to AI shifts the blame from strategic missteps or economic downturns to an external, seemingly unstoppable technological force, making it easier for companies to justify difficult decisions without acknowledging internal accountability.

But if executives like Bill McDermott are to be believed, the scale of job losses and the structural changes to the labor market could be staggering, regardless of the underlying causes. ServiceNow CEO Bill McDermott, for instance, told CNBC last week that he expects unemployment for new college graduates to reach over 30 percent in the coming years. Such a prediction, coming from a leader in the software and AI space, is not just an idle forecast but a potentially self-fulfilling prophecy if companies internalize this view and adjust their hiring practices accordingly. The implications for educational institutions, career counseling, and governmental support programs would be profound, necessitating a complete re-evaluation of curricula and vocational training.

In short, as Karpathy’s vibe-coded project, despite its rapid retraction and the subsequent clarifications, hints at, white-collar jobs are facing an existential threat unlike any seen in generations. The very professions that have long been considered stable, desirable, and intellectually stimulating are now at the forefront of AI’s transformative power. Conversely, more hands-on, often lower-paying, and physically demanding occupations could end up surviving the storm with greater resilience – a paradoxical outcome where traditional blue-collar work may prove more future-proof than many contemporary white-collar roles. This conclusion, while perhaps not definitive, certainly won’t be of much consolation to those in the midst of their post-secondary education, meticulously studying software development, accounting, or business administration, only to potentially enter a vastly altered and fiercely competitive job market.

The unfolding drama of AI’s impact on employment is a complex tapestry woven with technological progress, economic forces, corporate strategy, and human anxiety. While Karpathy’s chart was a fleeting, informal glimpse, it served as a powerful catalyst for a conversation that is only just beginning. The future of work, it seems, demands not just adaptability from the workforce, but also a critical and responsible approach from those developing and deploying these powerful AI technologies.

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