Andrej Karpathy dropped a job market visualization tool last week, and the data tells a story that the "AI is replacing everyone" narrative misses entirely. The tool pulls from public job posting data, categorizes roles by skill requirements, and visualizes trends over time. What it shows is not a simple story of AI eating jobs. It is a story of skill transformation that is moving faster than anyone expected.
What the Data Actually Shows
The headline number everyone is sharing: AI-related job postings are up 340% year-over-year. But that number hides the interesting detail. "AI-related" does not just mean ML engineers and data scientists. The fastest-growing category is "AI-adjacent" roles - positions that did not exist two years ago and require a hybrid of traditional skills plus AI fluency.
Examples from the tool's categorization:
- AI Operations Engineer. Not MLOps. Operations people who use AI agents to manage infrastructure. Think SRE meets AI automation.
- Agent Developer. People who build and configure AI agents for business processes. Not researchers. Practitioners who wire together models, tools, and workflows.
- AI QA Specialist. Testing AI-powered products. Not testing models (that is eval engineering). Testing the product experience when AI is a core component.
- Prompt Architect. Yes, this is a real job title now, and it pays well. Designing the prompt systems, tool schemas, and context management strategies that make AI products work reliably.
These roles pay well because they combine domain expertise with AI literacy, and that combination is genuinely scarce.
The Decline Is Real Too
Karpathy's tool also shows declines, and they are stark. Traditional data entry roles are down 65% year-over-year. Basic code-writing positions (implement this spec, no architecture decisions) are down 40%. First-line customer support roles are down 55%.
These are not future predictions. These are current hiring trends. Companies are not posting these jobs because AI already handles them, or because they are restructuring the remaining human roles to work alongside AI instead of doing the work AI now does.
The visualization makes this concrete in a way that abstract statistics do not. You can filter by city, industry, company size, and salary range, and the patterns hold. The shift is not concentrated in tech companies - it is happening across healthcare, finance, logistics, and manufacturing.
What This Means for Agent Builders
If you are in the AI agent space, the Karpathy data validates something you probably already felt: the demand for agent-related skills is growing faster than the supply of people who have them. "Can build and deploy AI agents" is becoming a career-defining skill.
But the tool also shows something less obvious: the companies hiring agent builders are not primarily AI companies. They are traditional businesses that realized they need AI agent capabilities. Banks, insurance companies, retailers, manufacturers. The job titles often do not even include "AI" - they say things like "Process Automation Lead" or "Digital Workforce Manager."
This means the market for agent expertise is much bigger than it appears if you only look at tech company job boards. The total addressable market for agent builders includes every company that has repetitive knowledge work, which is essentially every company.
The Geographic Spread
One of the most interesting features of Karpathy's tool is the geographic view. AI hiring is not just a San Francisco and New York phenomenon. The fastest growth in AI-adjacent roles is in second-tier cities - Austin, Denver, Atlanta, Chicago, Raleigh. Remote AI roles are also growing rapidly, which makes sense - if the work is building and configuring AI agents, it does not need to be done in a specific physical location.
For agent builders considering their next career move, the data suggests that opportunities outside the traditional tech hubs are growing faster and often come with better compensation-to-cost-of-living ratios.
The Tool Itself
The visualization is beautifully built (because of course it is - it is Karpathy). Interactive, fast, with sensible defaults and deep drill-down capabilities. It updates weekly with fresh data. If you make career decisions based on market data rather than Twitter vibes, bookmark it.
The underlying data is also available as an API, which means you could build agents that monitor job market trends and alert you to emerging opportunities. Several people in the comments are already doing this. The meta-irony of using AI agents to track the AI job market is not lost on me.