AI adoption among consumers is evolving, moving towards more sophisticated and knowledge-intensive applications, per a new analysis from the Anthropic Economic Index. Over eight months leading up to August 2025, usage patterns on Claude.ai have shown notable changes accompanying improvements in underlying model capabilities.
While computer and mathematical tasks still dominate overall usage at 36 per cent, there is sustained growth in knowledge-intensive fields. Educational instruction and library tasks surged from 9 per cent to 12 per cent of sampled conversations. Similarly, life, physical, and social science tasks increased from 6 per cent to 7 per cent.
Also Read: Anthropic index shows AI boom risks widening global inequality
This divergence suggests that AI usage is diffusing especially quickly among tasks involving knowledge synthesis and explanation, potentially because these tasks benefit more from Claude’s reasoning capabilities.
Programme creation overtakes debugging
A key insight into the growing reliability of AI models is seen within the dominant coding category. Tasks related to creating new code more than doubled, increasing by 4.5 percentage points (from 4.1 per cent to 8.6 per cent).
Concurrently, debugging and error correction tasks fell by 2.8 percentage points (from 16.1 per cent to 13.3 per cent). This net shift towards creation over fixing code suggests that models have become increasingly reliable, enabling users to accomplish more of their goals in a single exchange and spend less time fixing problems.
User confidence drives directive automation
The collaboration style between users and Claude is also shifting dramatically. The share of “directive” conversations, where users delegate complete tasks to Claude with minimal back-and-forth, jumped from 27 per cent to 39 per cent.
Also Read: Anthropic data shows businesses use AI to automate, not collaborate
This surge came primarily at the expense of “task iteration” and “learning” interactions. In this report iteration (V3), automation usage (now 49 per cent) has surpassed augmentation usage (45 per cent) for the first time. This shift signals growing user confidence in the AI’s ability to complete high-quality outputs on the first attempt, a form of learning-by-doing.
The post AI adoption evolves: Knowledge, confidence, and code creation soar appeared first on e27.