I made a job-level AI capability estimator by asking "Where is AI doing similar work today?"

Developer Alex A has created a novel tool that estimates AI capabilities across different occupations by asking a simple but powerful question: "Where is AI doing similar work today?" Built on top of Anthropic's Economic Index—which tracks actual AI usage by specific occupations on tasks from the O*NET database—the tool goes beyond mere adoption metrics to infer current AI capability. The core methodology involves creating vector embeddings for all task descriptions in the database and using semantic search to find the nearest similar tasks. If AI is being used for a semantically similar task in a different occupation, the tool updates the capability score for the original task, effectively cross-pollinating evidence of AI competence across job boundaries.

To ensure accuracy, Alex implemented a sanity-check layer using Claude Haiku to review scores and prevent misleading transfers—like a physical task inheriting a high score from a purely informational but semantically similar one. The tool, which visualizes data for over 800 occupations, was developed with assistance from Claude Opus 4.6. While the methodology has limitations (notably, it still shows 0% capability for some technical tasks like "Select programming languages" for Web Developers), it provides a more grounded, evidence-based view of AI's current reach into the workforce than speculative forecasts. The interactive result is available on Alex's LessWrong post, offering professionals a data-driven lens to assess automation risk and opportunity in their own roles.