Jun 16, 2026Economic ResearchAgentic coding and persistent returns to expertise

2026-06-11 · Source: Anthropic Research · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Data Science & Analytics · Depth: Advanced, extended

Summary

Anthropic's research, based on a privacy-preserving analysis of approximately 400,000 Claude Code sessions between October 2025 and April 2026, reveals key trends in agentic coding. The study found a clear division of labor: users typically make 70% of planning decisions, while Claude handles 80% of execution decisions. Domain expertise significantly amplifies Claude's output, with expert users prompting 12 actions and 3,200 words of output per turn, compared to novice users' 5 actions and 600 words. Notably, all major occupations achieve similar success rates on coding tasks as software engineers. Over the seven-month period, debugging sessions decreased by nearly half, and usage shifted towards more end-to-end agentic applications like deploying code and data analysis. The estimated economic value of tasks also increased by about 25% on average. Success rates are higher for more expert users, with verified success at 28-33% for intermediate/expert users versus 15% for novices.

Key takeaway

For Directors of AI/ML evaluating agentic coding tools, recognize that your team's domain expertise, not their coding proficiency, will dictate success. You should prioritize training on precise problem definition and effective agent instruction, as this amplifies agent output and improves task completion rates. This approach enables professionals across various occupations to successfully tackle complex technical work, broadening your team's capabilities beyond traditional software roles.

Key insights

Domain expertise, not coding skill, is the primary driver of success and agent utility in agentic coding.

Principles

• People define "what," agents define "how."
• Domain expertise amplifies agent output.
• Competence yields most benefits; mastery adds modestly.

Method

The study classified ~400,000 Claude Code sessions by work mode, decision attribution, and user expertise using privacy-preserving classifiers on transcripts and telemetry.

In practice

• Use agentic tools for diverse non-coding tasks.
• Focus on problem definition, not code syntax.
• Guide agents with precise domain context.

Topics

• Agentic Coding
• Human-AI Collaboration
• Domain Expertise
• AI Productivity
• Labor Market Shifts
• Claude Code

Best for: Executive, Research Scientist, AI Product Manager, AI Scientist, Director of AI/ML

Related on AIssential

• 10 Lessons for Agentic Coding — Agentic coding is transforming software development, requiring new approaches and skills.
• How I deleted 95% of my agent skills and got better results — Nick Nisi, WorkOS — Agent performance improves by enforcing actions with code, guiding with specific "gotchas," and rigorously measuring outcomes, not by comprehensive instruction.
• How I code with agents, without being 'technical' — AI agents enable non-technical users to build complex software by abstracting traditional coding into a guided, iterative process.
• How We Bet Against the Bitter Lesson — Agent Skills integrate human expert judgment and deterministic tools to create a new, efficient AI-human knowledge economy.
• Vibe Coding: The Complete Story (2015–2026) — The coding paradigm has shifted from human-driven to AI-agent-dominated, requiring new human oversight and expertise.
• Thoughts on coding agents — Coding agents fundamentally shift software engineering workflows, demanding explicit context provision and deliberate model-building from developers.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Anthropic Research.