AI & Work

How AI is reshaping the American workforce
Bureau of Labor Statistics data · scored by Gemini + Sonnet ensemble

Workers

—

Avg. Impact

—
job-weighted, 0–10 scale

High-exposure workers

—
in jobs scoring 7+

Wages at stake

—
annual wages in 7+ jobs

By impact level

Impact by pay

Impact by education

AI-Resilient Careers

High growth, low AI impact
→
created by chrona.nyc

Sources & Methodology

Data

Occupation data (employment, wages, education, growth outlook) comes from the Bureau of Labor Statistics Occupational Outlook Handbook (2024 edition), covering 342 occupations representing 160M+ American workers.

AI Impact Scoring

Each occupation is scored 0–10 for AI exposure by two independent LLMs — Gemini 3 Flash and Claude Sonnet 4.6 — then averaged. Both models analyze each job's core tasks, work environment, and required skills against the same calibrated rubric. The two models agreed within ±1 point on 91% of occupations.

  • 0–2 (Low): Physical, hands-on, or deeply interpersonal work
  • 3–5 (Moderate): Mix of AI-augmentable and distinctly human tasks
  • 6–7 (High): Primarily digital work with significant AI overlap
  • 8–10 (Very high): Core tasks can already be assisted by AI

Career Adjacency

"Paths Forward" suggestions use BLS Similar Occupations data (expert-curated relationships between jobs), weighted by education match, pay similarity, and growth outlook.

Model Agreement

The two models agreed within ±1 point on 91% of occupations. Where they disagree, Sonnet tends to score knowledge work lower, giving more weight to human judgment and physical presence. Gemini leans toward "if it's digital, it's exposed."

Biggest disagreements:

  • Sports officials: Sonnet 2, Gemini 5 — how digital is refereeing?
  • Aerospace engineers: Sonnet 6, Gemini 8 — hardware vs. design software
  • Air traffic controllers: Sonnet 5, Gemini 7 — real-time safety judgment
  • Actors: Sonnet 5, Gemini 7 — physical performance vs. digital media

Scoring Prompt

Both models receive the same calibrated rubric with anchor examples at each level. The full prompt and scoring pipeline are in the source code.

Limitations

AI impact scores reflect a single model's assessment and should be treated as directional estimates, not precise predictions. AI capabilities are evolving rapidly — scores may shift as the technology matures. Growth projections are BLS 10-year estimates (2023–2033).

Credits

Originally created by Andrej Karpathy. Extended by Chrona with career path analysis, interactive filtering, and additional analytics.

Source code: github.com/chrona-nyc/jobs

AI-Resilient Careers

High growth outlook and low AI exposure — jobs where human skills are in demand.