【AI前沿】Who Uses Claude Code? Latest User Portrait: Working 20 Hours a Week on Average

AI NEWSLatest AI NewsArticleWho Uses Claude Code? Latest User Portrait: Working 20 Hours a Week on AveragePublished in Latest AI NewsTime :Jun 18, 2026Read :4minuteAs AI coding assistants gradually permeate developers’ daily workflows, more and more programmers are exploring the boundaries of human-machine collaboration. Recently, Anthropic released a detailed analysis report, based on an in-depth exploration of about 235,000 users and 400,000 Claude Code sessions, outlining the true picture of this group of “AI coding pioneers.“Data shows that these deep users of AI demonstrate remarkable engagement, spending an average of 20 hours per week on Claude Code. In terms of occupational distribution, professionals in computer science and mathematics dominate, while professionals in business finance, art design, management, and research are also becoming important users of AI coding tools.In terms of specific human-machine division of labor, the collaboration model has become quite seamless: when a task starts, human developers play the role of “pilot,” responsible for about 70% of planning decisions, determining “what to do”; while Claude becomes the “executor,” taking on about 80% of the specific operations, handling “how to do it.” This efficient division of labor significantly improves programming efficiency.Looking at the work content, AI coding tools have already penetrated every aspect of development. Among them, code modification tasks account for 26%, followed by 25% of code writing work, with both together accounting for more than half. In addition, AI also performs well in handling software operations, writing documents, and creating presentations.Anthropic’s research also found an interesting phenomenon: the depth of AI usage is closely related to the user’s professional level. By analyzing indicators such as the need for prompt verification and error correction frequency, researchers divided users into five ability levels. Comparing them, novice users (level 1) typically produce about 5 actions and 600 words of output per interaction; while senior experts (level 5) can better utilize the model, generating twice as many actions as the former and up to an astonishing 3,200 words of output, which is five times that of the former.This report not only reflects the current market enthusiasm for AI-assisted programming but also reveals a trend: as AI tools continue to evolve, developers are not just using tools, but also continuously enhancing the depth of human-machine collaboration to cope with increasingly complex engineering challenges.