Anthropic says Claude is reshaping its own engineering work

Anthropic has published an unusually detailed look at how its own engineers and researchers use Claude at work, turning the company into a case study for AI-assisted software development.

The company surveyed 132 technical employees, interviewed 53 people and analyzed Claude Code usage. Employees reported using Claude in 59% of their work, up from 28% a year earlier, and estimated an average 50% productivity boost. Anthropic also found that 27% of Claude-assisted work involved tasks that probably would not have been done otherwise, such as small internal tools, exploratory analysis and quality-of-life fixes.

The report is not a clean productivity benchmark. Anthropic is measuring itself, the survey is self-reported and its employees have unusually early access to frontier tools. But the details are useful because they show what may spread to other technical teams: engineers using AI first for debugging and code understanding, then gradually for feature work, refactors and unfamiliar parts of the stack.

Anthropic also names the trade-offs. Employees said Claude can reduce dependence on colleagues for routine questions, weaken some deep technical practice and make supervision harder when output becomes cheap and fast. The practical takeaway is not just that coding agents save time, but that teams need new habits for review, mentorship and deciding which work should remain human-led.