Andrej Karpathy, the former Tesla and OpenAI researcher, built an AI-powered job map by scraping federal labor data, and the results point to a clear pattern: occupations with median pay above $100,000 per year consistently rank highest for what the tool calls “exposure,” a composite measure blending salary, projected growth, and the degree to which AI intersects with the work. The visualization draws entirely from the U.S. Bureau of Labor Statistics, turning dry government tables into an interactive tool that lets workers and employers see which roles carry the most economic weight in an AI-influenced economy. The finding raises a pointed question about whether high-wage knowledge work is best positioned to absorb AI or most vulnerable to disruption by it.
What the Job Map Actually Measures
Karpathy’s tool pulls from the Occupational Outlook Handbook, a BLS publication that catalogs hundreds of occupations by their definitions, typical duties, work environments, education requirements, median pay, and projected growth. By layering an AI-derived “exposure” score on top of these government baselines, the map ranks jobs not by a single variable but by how multiple labor-market signals stack up together.
The distinction matters because “exposure” here is not synonymous with “risk of replacement.” A software developer earning well above $100,000 scores high on the exposure index partly because the role intersects heavily with AI tooling, but also because demand projections remain strong. A low-wage retail position, by contrast, may face automation pressure yet scores lower because its pay and growth outlook pull the composite metric down. The map, in other words, rewards concentration of economic signals rather than measuring threat alone.
BLS Projections Behind the Numbers
The credibility of any visualization depends on the quality of its inputs. Karpathy’s map inherits its baseline from BLS occupational projections, which include downloadable tables, definitions, and crosswalks that connect occupation codes to real workforce counts. These tables form the backbone of the Outlook Handbook and are the same data that federal agencies, universities, and workforce boards use when planning training programs or allocating grants.
The methodology behind those projections is documented in the BLS employment methods, which describe a multi-stage economic model. The model draws on industry output estimates from the Bureau of Economic Analysis, which publishes detailed national accounts, manufacturing data from the Census Bureau’s manufactures survey, broader industry benchmarks from the economic census, and energy price assumptions from the Energy Information Administration. Each of those inputs feeds a demand-side forecast that BLS then translates into occupation-level job counts. When Karpathy scraped the Handbook’s output, he was pulling from a pipeline that stretches across multiple federal statistical agencies.
Why $100K Roles Cluster at the Top
The pattern the map reveals is not random. High-paying occupations tend to require advanced education, sit in sectors with strong projected hiring, and involve tasks that overlap with AI capabilities, whether that means writing code, analyzing data, or managing complex systems. All three factors push the exposure score upward simultaneously.
A Labor Department post published alongside BLS projection releases offers supporting context: STEM occupations are projected to grow significantly faster than the average for all occupations, and their median wages sit well above non-STEM counterparts. That wage and growth premium is exactly what Karpathy’s composite metric captures. Roles in software development, information security, data science, and health-care technology all share the combination of high pay, fast growth, and deep AI relevance that places them at the top of the map.
The concentration at the $100K threshold is not a coincidence of the scoring formula. It reflects a structural reality in the U.S. labor market: the occupations most intertwined with emerging technology also happen to command the highest compensation, creating a feedback loop where investment in AI tools flows disproportionately toward already well-paid fields.
What the Map Misses
Most coverage of Karpathy’s tool has treated the exposure score as a straightforward ranking of AI impact. That reading oversimplifies the picture. The BLS projections that feed the map are baseline forecasts, not predictions. They assume a continuation of recent trends in technology adoption, labor force participation, and economic growth. They do not model sudden shifts such as a recession, a regulatory crackdown on AI deployment, or a breakthrough that makes an entirely new class of jobs obsolete overnight.
The map also inherits the geographic blindness of its source data. The Occupational Outlook Handbook describes work environments in general terms, noting whether a role is typically office-based, field-based, or remote-eligible. But it does not break out job counts or wages by metro area. A $100K software role in San Francisco carries very different purchasing power than the same title in Tulsa, yet both receive the same exposure score. For workers trying to make career decisions based on the tool, that gap between national medians and local reality is significant.
There is also a missing piece in the public record. Karpathy has not released detailed documentation of how the AI-derived exposure layer was constructed, what model generated the scores, or how the weighting between pay, growth, and AI relevance was calibrated. The BLS inputs are transparent and reproducible through Bureau of Economic Analysis data tables and BLS data tools, but the proprietary overlay remains a black box. That matters because the exposure score is the map’s entire value proposition. Without methodological transparency, users are trusting an opaque algorithm applied on top of trustworthy government data.
Practical Takeaways for Job Seekers
Despite its limitations, the map does something useful: it collapses complex labor-market information into a visual hierarchy that is easier to scan than a stack of spreadsheets. For an individual deciding where to invest time and training, the high-exposure cluster highlights roles that combine strong pay, solid growth, and meaningful interaction with AI systems.
One takeaway is that AI fluency is becoming a baseline expectation in many of the best-compensated fields, not just in obvious tech roles. Occupations like marketing management, financial analysis, and certain health-care specialties may not be labeled as “AI jobs,” yet they increasingly rely on data-driven tools, predictive models, and automated workflows. Workers in these areas can use the map as a prompt to identify where AI is already embedded in their daily tasks and where upskilling (through coursework, certifications, or on-the-job experimentation) could raise their value.
Another lesson is that exposure cuts both ways. A high score signals opportunity but also volatility. If AI tools dramatically increase the productivity of software engineers, for example, employers might demand fewer engineers in the long run, even as the remaining roles become more complex and better paid. Job seekers should read high exposure as a cue to focus on skills that are complementary to automation (problem framing, cross-functional communication, ethical judgment) rather than on narrow tasks that are easiest for machines to replicate.
The map can also help people in lower-paying roles think about adjacent moves. A customer service representative may see that their current occupation ranks low on exposure, but nearby roles in technical support or sales engineering sit much higher. Because the underlying BLS data describe education levels and typical pathways, workers can pair the visualization with official occupation profiles to identify realistic steps, such as earning a certification or learning a specific software platform, that move them toward better-paid, more AI-resilient work.
Implications for Employers and Policymakers
For employers, the clustering of high exposure around $100,000-plus roles underscores that AI is not just a cost-cutting tool for low-wage work. It is reshaping the most expensive segments of the payroll. Companies relying heavily on software development, analytics, or technical management should anticipate that these positions will continue to evolve quickly as AI capabilities advance. Strategic workforce planning will need to address not only how many of these roles are needed, but how their skill requirements are shifting.
Policymakers, meanwhile, can read the map as an early signal of where training and education resources might have the highest leverage. Because the underlying projections are grounded in BLS modeling and Census and BEA inputs, the high-exposure cluster aligns with areas where the broader economy expects sustained demand. Directing scholarships, apprenticeships, and community-college programs toward these occupations could help more workers access the wage and growth premium that the visualization highlights.
At the same time, the absence of geography and the opacity of the AI scoring argue for caution in using the map as a formal planning tool. Regional workforce boards and state agencies will still need localized data on wages, vacancies, and employer needs. The map is best viewed as a starting point, a way to prioritize which occupations deserve deeper, region-specific analysis, rather than as a definitive ranking of where AI will hit hardest.
Reading the Map Without Overreading It
Karpathy’s project succeeds in making federal labor statistics feel immediate and legible to a broader audience. It surfaces a real and important pattern: the jobs that pay the most and are projected to grow the fastest are also the ones most intertwined with AI. But the tool is a lens, not an oracle. Its exposure scores rest on a mix of transparent government baselines and opaque algorithmic judgments, and they abstract away regional nuance and future shocks.
For workers, the most productive response is not to panic about AI “coming for” high-wage roles, nor to assume that a high exposure score guarantees security. It is to use the map as one input among many, combining it with personal interests, local labor conditions, and a realistic assessment of how comfortable they are working alongside rapidly evolving technology. In that sense, the visualization does its job: it invites people to look squarely at how AI and economic value are converging, and to make more informed choices about where they want to stand as that convergence accelerates.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
