AI coding agent usage doubles, experts get much more done

Anthropic just released a big study based on 400,000 Claude Code sessions from 235,000 users. They found AI coding agent usage has doubled in recent months, with users spending about 20 hours per week on average. The data was collected using their privacy-focused tool Clio.

About 70% of sessions came from identifiable jobs, mostly computer and math roles, followed by business, design, and management. In a typical session, humans handle 70% of planning decisions, while Claude handles 80% of execution. Code editing and writing make up over half of all tasks.

Anthropic also ranked users by expertise from 1 to 5. Expert users (level 5) get about 10 actions and 3,200 words per prompt, which is 2x more actions and 5x more output than beginners (level 1). This shows how much domain knowledge boosts AI collaboration efficiency.