Most professional use of generative AI augments workers' tasks rather than automating them.
Assessed first pass
Budget-constrained first pass. Left atomic (no subclaims); assessed inconclusive (credence 0.45) from Anthropic Economic Index data: consumer chat ~52-53% augmentation, business API ~77% automation. No instances recorded due to budget; a later pass should record Economic Index instances and examine non-Anthropic usage data (Microsoft Copilot, OpenAI usage studies).
Assessed Inconclusive
verdict confidence 0.65 · credence 0.45
The best large-scale usage evidence on this question comes from the Anthropic Economic Index, which classifies conversations with Claude into interaction modes grouped as augmentation (learning, task iteration, validation) or automation (directive and feedback-loop delegation). On the consumer Claude.ai product, augmentation was about 57 percent in early 2025, fell to 47 percent against 49 percent automation in August 2025, recovered to about 52 percent in November 2025, and rose slightly again in February 2026. On the first-party business API, however, about 77 percent of traffic in 2025 showed automation patterns and only about 12 percent augmentation, and coding work has since been migrating from the chat product to automated API workflows such as Claude Code. Whether "most professional use" augments therefore depends on which channel is counted as professional use. Consumer chat traffic includes large shares of personal and coursework use, and its work-related portion is close to evenly split. Programmatic business use, which arguably is the most clearly professional, is strongly automation-dominated. Weighting by volume of work done rather than by conversation count would likely push the balance further toward automation. A separate difficulty is definitional: a conversation classified as "directive" automates a sub-task inside a job, which is not the same as automating the worker's task in the labor-economics sense; worker surveys commonly report that people use AI to assist rather than replace their work, which favors the claim under a looser reading. The claim is therefore neither established nor refuted. It holds roughly for interactive chat use by workers and fails for business API deployments. Data from other providers (for example Microsoft and OpenAI usage studies) and a usage-weighted definition of "professional use" would be needed to settle it.
Claim entered the graph