Anthropic research: domain expertise—not coding background—predicts success with Claude Code, based on 400,000 real sessions

Anthropic published research on June 16 showing that the single strongest predictor of success with Claude Code is domain expertise in the task at hand — not whether a user has a background in software engineering. The analysis covers roughly 400,000 interactive Claude Code sessions from approximately 235,000 users between October 2025 and April 2026.

What's new

The research draws on Anthropic's own usage data to map how different types of users work with Claude Code. Key findings:

• Expertise drives output. Expert users triggered approximately 12 autonomous Claude actions per prompt, versus 5 for novice users. As the paper states: "The greater domain expertise a person brings to a session, the more work Claude does per instruction."
• Non-engineers succeed at the same rate as engineers. "On coding tasks, every major occupation succeeds at nearly the same rate as software engineers, on average." Success was measured by verifiable outcomes — passing tests, committed work, or completed deliverables.
• Success rates scale with expertise. Novice sessions achieved verified success 15% of the time; intermediate and expert sessions reached 28–33%.
• Task value is rising. "Over those seven months, the value of the typical task rose in almost every kind of work — about 25% on average." Building and operating tasks rose 32–43%.
• Clear division of labor. Users make roughly 70% of planning decisions (what to build), while Claude handles about 80% of execution decisions (how to build it).

The 400,000 sessions span 235,000 users and break down as: 56% writing, fixing, or testing code; 17% operating software; 13% producing analysis or prose.

Context

Claude Code launched in late 2025 as an agentic coding tool that autonomously completes multi-step software tasks from a terminal or IDE. The study period — October 2025 through April 2026 — covers major model upgrades including Claude Sonnet 4.6, Opus 4.7, Opus 4.8, and Fable 5.

The research was designed partly to characterize who actually uses Claude Code and how. Prior assumptions about AI coding assistants generally placed programming knowledge at the center; this real-world data challenges that directly.

Why it matters

The finding that domain expertise beats coding background has direct implications for enterprise AI deployment. Pairing a domain expert who understands the problem deeply with an agent that handles execution yields substantially better results than deploying the tool to a novice coder who knows syntax but not the underlying domain.

As Anthropic frames it: "Success is determined by how well a person understands the problem they are trying to solve, not whether you're trained in coding."

This reframes the ROI calculation for businesses deploying AI coding tools. The leverage is not in retraining non-technical staff to code, but in connecting domain experts directly to capable agents. The 25% average rise in estimated task value over seven months suggests that as models improve, this dynamic will only strengthen — more value per session, more broadly distributed across professions.