Irreducibly Human

The AI Exposure Explorer

Pick two occupations to compare them two ways: how their employment has moved since 2018, set against major AI milestones (Chart 1), and the human abilities each job leans on (Chart 2). The lens is the Irreducibly Human question of which abilities are hardest for AI to replicate — the seven-tier taxonomy explains the idea behind the tool.

  • Choose two occupations with the selector below (search by role name or SOC code).
  • In Chart 2, switch between COMPARISON and TOP DIFFERENCES, browse the ability categories, and open the Glossary for any unfamiliar term.
  • Hover any point for its exact values.
Select two occupations to compare
Occupation 1
Occupation 2

You can search any occupation. Not all have BLS employment data: when it's missing, Chart 1 omits that line (and says so) while Chart 2 still compares abilities.

Chart 1 — Employment

Hover anywhere on the chart to see the exact index value and employment count for that year. Hover a milestone label below the axis for the date and description of that AI development milestone.
Software Developers wasn't tracked separately by BLS in 2019-2020: it was folded into a combined code with QA Analysts (SOC 15-1256), so data for Software Developers alone isn't available for those years.
Legend
Software Developers
Computer Programmers
Data gap (intentional, not missing)
2018 baseline (index = 100)
What the index means

Each occupation's employment is indexed to 2018 = 100, so values compare directly: 48 means employment fell to 48% of its 2018 level; 126 means it grew to 126%.

The milestone timeline

Labels below the axis mark major AI milestones, like model releases and coding-assistant launches, so employment shifts can be compared against AI capability jumps.

Built by Abisha Vadukoot, Milivoje (Mickey) Davidovic, and Nik Bear Brown, with data from O*NET and the U.S. Bureau of Labor Statistics (BLS).

Chart 2 — Abilities

Pattern, language, recall, and calculation. Machines are strongest here (Tier 1), so high demand does not mean hard to automate.
  • Each dot is an occupation's rating for that ability. The short vertical tick is the average across all occupations in the workforce. Hover a dot for its confidence interval and sample size.
  • Ratings use O*NET's Level scale (0–7): how much of the ability the job requires. The shaded band near 0 marks abilities the job does not require.
  • Tap an ability name to open the glossary and see what the term means.
Software DevelopersComputer ProgrammersWorkforce Average
Where does each occupation need more or less of a given human ability?
Machines are now superhuman at the pattern recognition, recall, and calculation that make up much of O*NET's Cognitive abilities. Use this profile to see which abilities a job leans on, not as a ranking of what is safe from automation: the capacities hardest for AI to replicate (judgment, causal reasoning, and embodied or collective know-how) sit largely above what these measures capture.
Built by Abisha Vadukoot, Milivoje (Mickey) Davidovic, and Nik Bear Brown, with data from O*NET and the U.S. Bureau of Labor Statistics (BLS).