About this tool

How it was built, where the data comes from, and what the scores mean.

What this is

This tool lets you look up any of 342 US occupations and see how exposed that role is to AI disruption, scored from 1 (minimal exposure) to 10 (maximum exposure). It covers 143 million jobs across the US economy.

It is a research tool, not a prediction. A high score does not mean a job will disappear. It means the nature of that work is likely to change significantly as AI capabilities advance.

Data sources

Three datasets power this tool:

  • 1.
    Occupational data — from the US Bureau of Labor Statistics Occupational Outlook Handbook. This provides job titles, employment numbers, median pay, growth outlook, and education requirements for each occupation. This is the standard reference used by the US government, researchers, and economists.
  • 2.
    AI exposure scores — generated using a methodology developed by Andrej Karpathy, co-founder of OpenAI and former director of AI at Tesla. Each occupation's BLS description was scored by a large language model on a 0–10 scale for “Digital AI Exposure” — how much current AI capabilities will reshape that occupation. The full scoring prompt is published on Karpathy's site.
  • 3.
    Job title search — powered by the O*NET Alternate Titles database from the US Department of Labor. This maps over 15,000 common “lay” job titles to their official BLS occupation categories — so searching “management consultant” finds “Management analysts,” and “doctor” finds “Physicians and surgeons.”

What we did and didn't change

We did not modify any data.The exposure scores, rationales, employment numbers, pay figures, and growth outlooks are exactly as produced by Karpathy's analysis of BLS data. We are presenting the data, not creating it.

What we added: the search interface, the O*NET title mapping (to make occupations findable by common job titles), the contextual guidance text (“Where to focus”), and the visual design. The guidance text represents our interpretation of what the scores might mean in practice — it is editorial, not data.

Scoring methodology

The AI exposure score measures how much current digital AI capabilities will reshape each occupation. The key signal is whether a job's work product is fundamentally digital:

  • 1–3Mostly physical or interpersonal work. AI helps with peripheral tasks but doesn't touch the core role.
  • 4–5A mix of physical and knowledge work. AI meaningfully assists with information processing.
  • 6–7Predominantly knowledge work. AI tools are already making workers substantially more productive.
  • 8–10Almost entirely digital work in domains where AI is advancing rapidly. The occupation faces major restructuring.

The full scoring prompt, including calibration anchors and example occupations, is available on Karpathy's site.

Important limitations

  • →These are estimates, not predictions. The scores reflect rough LLM-generated assessments of how much AI could reshape each occupation. They are not peer-reviewed research.
  • →High exposure ≠ job loss. Software developers score 9/10 because AI is transforming their work — but demand for software developers continues to grow. Exposure measures restructuring, not replacement.
  • →Methodological circularity. Using an LLM to score jobs for LLM replaceability introduces inherent bias. The scores reflect what AI systems believe they can do, which may overestimate or underestimate actual impact.
  • →US-centric data. The BLS data covers the US labor market. Job structures, regulations, and AI adoption rates vary significantly by country.
  • →Point-in-time snapshot. AI capabilities are evolving rapidly. These scores reflect current digital AI. Physical AI (robotics, autonomous systems) will change the picture for many low-scoring occupations.
  • →342 of ~800+ detailed occupations. The BLS Occupational Outlook Handbook covers a subset of all US jobs. Some roles may not appear in the dataset.

Who built this

This tool was built by Lion Strategy, a strategy consultancy focused on AI, digital assets, and frontier technology. We built it because we believe people deserve clear, accessible information about how AI is reshaping work — without hype, fear, or a sales pitch.

The source data is freely available. The BLS Occupational Outlook Handbook is a public resource. Karpathy's methodology and scoring prompt are published openly. The O*NET database is maintained by the US Department of Labor.

Missing a role?

If you don't see your role, please email howexposed@gmail.com to let us know.