Title: The Thermodynamics of AI Adoption: Cognitive Divergence Theory and the Anthropic Economic Index
Tags: validation-cost, work-and-productivity, skill-divide
---
This short research note compares the January 2026 report of the Anthropic Economic Index with the author's papers of November and December 2025. The report, as the note reads it, lowered the projected productivity gain from AI from 1.8% to between 1.0 and 1.2% once the cost of checking outputs was included, and recorded a shift in usage from automation towards augmentation.

The note maps three findings onto three earlier concepts. The lower outlook is matched with the psychological ceiling: a tool cannot exceed the competence of the person who validates its output. The hidden validation cost is matched with **cognitive entropy**, called here a "thermodynamic tax" paid when a task is delegated without the competence to verify it. The move from automation (48% to 45%) to augmentation (47% to 52%) is read as the collapse of **False Cognitive Power Transfer**, the illusion that speed of generation is the same as skill. A reported correlation of 0.93 between the quality of prompts and the quality of outputs is taken as support for the claim that AI amplifies existing ability.

A discussion section addresses the distance between a leverage of 71 times for experts and an average gain of about 1%. Leverage is defined as the ratio between the speed at which an expert can validate text and the speed of manual writing. A population split into Passengers (70%), Operators (25%) and Architects (5%), each with an assumed leverage, gives a weighted figure that, applied to the share of tasks open to AI, comes close to the reported gain.

The note concludes that the data confirm the theory and that forecasts should include the cost of verification.