While the ability of AI tools to reshape entire human occupations remains a subject of heated debate, a sobering reality is starting to settle in. The echoes of past industrial revolutions, where technologies like the steam engine irrevocably altered the labor landscape, are now resonating in the digital age, prompting urgent questions about the future of work.

Big tech companies are laying off workers in the thousands, with CEOs expecting the worst and predicting soaring unemployment rates among college graduates – all while gleefully cutting costs at their companies and not looking back. This confluence of technological advancement and corporate restructuring has intensified public and professional interest in understanding which roles are most vulnerable to the rapid proliferation of artificial intelligence.

Amidst this atmosphere of uncertainty, Andrej Karpathy – an OpenAI cofounder, former AI executive at Tesla, and inventor of “vibe coding” – recently ignited a fresh wave of discussion. Leveraging Bureau of Labor Statistics (BLS) data and a significant application of AI, Karpathy developed an interactive chart designed to rate jobs on a scale of zero to ten. On this scale, zero signified minimal exposure to AI-driven displacement, while ten indicated maximum susceptibility.

His project, though quickly withdrawn, captured immediate attention. As Fortune notes, the interactive chart drew considerable scrutiny before Karpathy, perhaps overwhelmed by the reaction, opted to pull it down. An archived version, however, still offers a glimpse into his initial findings and methodology.

Karpathy swiftly clarified his intentions and the nature of his work. “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.” He emphasized that the project was intended as a tool for exploration, not a definitive forecast.

However, the rapid and widespread interpretation of his chart led to its removal. “It’s been wildly misinterpreted (which I should have anticipated even despite the readme docs), so I took it down,” he added, highlighting the challenges of communicating complex AI-driven analyses to a broad audience.

Further elaborating on the core metric, 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.” He stressed that the metric was a simplified proxy for digital task susceptibility, not a holistic economic model for job displacement. “People are sensationalizing the visualization tool and putting words in my mouth,” he concluded, underscoring the gap between technical intent and public perception.

While Karpathy’s findings should indeed be taken with a heavy dose of skepticism – AI models still grapple with widespread hallucinations, and Karpathy himself advocates for “vibe coding” primarily for rapid iterations and “throwaway weekend projects” – the general patterns revealed by his data resonate with other independent analyses. His interactive chart, even in its brief public existence, suggested that occupations requiring significant physical presence, manual dexterity, or complex human interaction might be relatively insulated from AI’s immediate impact. This included roles such as construction laborers, janitors, electricians, barbers, and bartenders, which largely fell into the lower end of his “exposure” scale.

Conversely, jobs characterized by repetitive cognitive tasks, data processing, and digital communication appeared to be at the highest risk. Accountants, office clerks, customer service representatives, and software developers were identified as potentially among the hardest hit. This distinction between “physical” and “digital” tasks, and between human-centric and data-centric roles, forms a critical lens through which to understand AI’s potential influence on the labor market.

This broad conclusion is not unique to Karpathy’s “vibe-coded” exploration. AI company Anthropic’s own investigation into the matter yielded strikingly similar insights. Earlier this month, Anthropic released its latest findings about the “labor market impacts of AI,” a more formal and extensive research endeavor. Their researchers independently identified computer programmers, customer service representatives, data entry keyers, medical record specialists, and market research analysts as occupations with the highest risk, or “most exposed,” to AI-driven automation and augmentation. The convergence of these findings, despite different methodologies and underlying models, strengthens the argument that certain white-collar, information-heavy roles are indeed facing significant transformative pressures.

However, the question of whether employment levels are on the precipice of a catastrophic decline due to the rampant deployment of generative AI in the workplace remains a subject of intense debate. Anthropic’s report itself offers a crucial nuance, pointing out that “AI is far from reaching its theoretical capability” and that “actual coverage remains a fraction of what’s feasible.” This suggests that while the *potential* for AI to displace jobs is vast, the *real-world implementation* and its resulting impact will likely be a more gradual and complex process, influenced by economic factors, regulatory frameworks, and societal adoption rates.

Moreover, the narrative of AI-driven job losses has faced scrutiny, particularly concerning the motivations behind current tech sector layoffs. Many critics argue that tech leaders, in justifying mass layoffs by citing AI, are potentially deflecting attention from deeper issues such as corporate bloat and past overhiring. During the pandemic-fueled tech boom, many companies expanded rapidly, leading to what some now describe as unsustainable staffing levels. In this view, AI serves as a convenient, technologically advanced justification for necessary, but painful, corporate restructuring rather than being the sole, direct cause of widespread job displacement.

Despite these counterarguments, the pronouncements from some executives paint a grim picture of future employment. ServiceNow CEO Bill McDermott, for instance, told CNBC last week that he anticipates unemployment for new college graduates could soar to over 30 percent in the coming years. Such predictions, if they materialize, would represent an unprecedented disruption to the labor market and highlight the profound challenges facing educational institutions and policymakers.

In short, as Karpathy’s vibe-coded project hints at, and as more rigorous studies from Anthropic confirm, white-collar jobs are indeed facing an existential threat of transformation or partial automation. Roles that were once considered secure and intellectually demanding are now susceptible to automation, especially those involving data analysis, administrative tasks, and even certain aspects of software development. Conversely, more hands-on, physically demanding, or inherently human-centric occupations, often categorized as blue-collar or service-oriented, could paradoxically end up surviving the storm with greater resilience.

This conclusion, while potentially offering some solace to those in manual trades, won’t be of much consolation to individuals currently pursuing or having recently completed post-secondary education in fields like software development, accounting, or business administration. It necessitates a critical re-evaluation of educational curricula, a strong emphasis on skills that complement AI rather than compete with it (such as critical thinking, creativity, emotional intelligence, and complex problem-solving), and a societal focus on reskilling and lifelong learning initiatives.

The transition driven by AI is not merely about job elimination but also about job evolution and creation. New roles centered around AI ethics, prompt engineering, AI system oversight, and human-AI collaboration are already emerging. The historical precedent of technological shifts, from the agricultural revolution to the digital age, demonstrates that while specific jobs disappear, new ones often emerge, albeit requiring different skill sets. The challenge lies in managing this transition equitably and ensuring that the benefits of AI are broadly shared, rather than exacerbating existing inequalities.

The discourse surrounding AI’s impact on employment is complex and multi-faceted. It involves not just technological capabilities but also economic incentives, corporate strategies, and human adaptability. While Karpathy’s quick experiment provided a provocative, albeit quickly retracted, snapshot, it served as a powerful reminder of the profound questions we face. The future of work will demand continuous learning, adaptability, and a nuanced understanding of how humans and intelligent machines can best collaborate to drive innovation and prosperity.

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