AIAcademy · AIAcademy · 2026-05-16
Three reports in five months, from the same anonymized Claude.ai conversation corpus. Read in series they constitute the most concrete picture of measurable AI workplace adoption we have.
January 2026 — "Economic Primitives." The first Index release framed AI use through ONET task taxonomies: which occupational tasks Claude is actually being asked to perform, and at what frequency. The headline finding was that ~37% of Claude conversations map cleanly to a single ONET work activity, concentrated in software development, technical writing, and analytical reasoning. Augmentation outpaced automation in the conversation mix by roughly 2:1 — most people are iterating with Claude rather than handing it whole jobs.
March 2026 — "Learning Curves." The follow-up looked at the shape of adoption over time within a cohort. Three findings stuck: (1) heavy users plateau quickly — most of the gain in conversation depth happens in the first six weeks; (2) "expert-level" prompting behaviors (system messages, structured outputs, multi-turn refinement) remain rare even among power users — under 8% by their internal classifier; (3) industry mix shifts seasonally, with education spiking each semester and finance compressing into quarter-end weeks.
May 2026 — the 81k-interview deep dive. The most recent release pairs the conversation-corpus telemetry with structured interviews of 81,000 Claude.ai users across 130 countries. The interview side surfaces what the telemetry can't: why people stop using Claude (mostly trust and accuracy, not capability ceilings), what they wish it did (memory, persistence, integration with the rest of their working life), and how use varies by national context. The associated blog post draws the through-line across all three reports.