Anthropic Study Finds AI Enables 27% New Work, Boosts Productivity by 50%

TL;DR: Anthropic’s internal study shows AI tools let staff take on 27% of tasks they would not have attempted, driving an average 50% productivity gain.

Context Enterprises have measured AI impact mainly by speed gains on existing work. Recent data suggests a broader effect: AI can make previously impractical tasks viable, expanding the scope of what employees accomplish.

Key Facts Anthropic surveyed engineers and researchers, analyzed 200,000 Claude Code transcripts, and conducted 53 in‑depth interviews. Employees now use Claude for 60% of their daily work, up from 28% a year earlier. Reported productivity gains rose from 20% to roughly 50% over the same period. The study found that 27% of AI‑assisted output came from tasks employees would not have tried without the model, indicating AI is unlocking new work rather than merely accelerating old work.

The shift is reflected in task complexity. The average number of consecutive tool calls Claude completed without human input doubled from about 10 to 21. Tasks involving new feature implementation grew from 14% to 37%, while engineers used AI to build dashboards, revive stalled projects, and run exploratory research that previously lacked a cost justification.

OpenAI’s enterprise research echoes these findings: 75% of surveyed workers said AI enabled them to complete tasks they could not perform before. Across the broader market, 39% of firms are reinvesting AI‑driven productivity gains into research and development, suggesting the effect extends beyond individual efficiency.

What It Means AI is moving from a speed‑up tool to a capability enhancer, allowing employees to cross the viability threshold for low‑value‑time work. Companies that can integrate AI into new problem spaces may see outsized returns, while those hindered by talent shortages or organizational readiness risk lagging behind. Anthropic plans to broaden its research beyond engineers to gauge AI’s impact across functions, a move that could clarify how widespread the “new work” effect will be.

Watch for follow‑up studies in 2026 that compare AI‑enabled task expansion across industries and assess whether the productivity boost sustains as adoption widens.