Anthropic study of 400,000 Claude Code sessions: domain expertise matters more than coding skill

What happened

On June 22 Anthropic released "Agentic coding and persistent returns to expertise" — an analysis of 400,000 Claude Code sessions from ~235,000 users between October 2025 and April 2026.

Context and impact

It flips the "LLMs level the field" thesis. Experts pull more work per instruction, know what they want more precisely, and finish with a verified result more often. For teams, that means the return on training senior staff actually grows with agentic tooling, not shrinks.

Details

• Verified success: experts hit it 2x more often than novices
• Expert session: ~12 actions and ~3,200 words per prompt
• Novice session: ~5 actions and ~600 words per prompt
• Sample: 400,000 sessions, 235,000 unique users
• Window: October 2025 – April 2026
• The decisive factor is domain expertise (finance, law, biology...) not just coding background