Anthropic Reports Claude Now Writes Over 80 Percent of Its Own Production Code
Anthropic has published internal metrics showing its Claude AI model is responsible for more than 80 percent of the code merged into the company's production codebase, and that engineer output has risen sharply since the practice was adopted. The figures are self-reported and have not been independently verified.
Anthropic has published internal metrics indicating that its Claude AI model now writes more than 80 percent of the code merged into the company's own production codebase, according to a report published on 21 September 2026.
The company also reported that engineer output has risen sharply since the AI-assisted coding practice was adopted, though no specific productivity figures were disclosed in the available evidence.
The metrics are self-reported and internal to Anthropic. The source note accompanying the figures advises that technology leaders considering similar approaches "should validate any comparable gains by measuring defect rates, code review time, maintainability, and incident outcomes in their own environments." No independent third-party verification of the figures has been reported.
The disclosure is notable as an example of an AI company using its own model extensively in its software development workflow, a practice sometimes referred to as "dogfooding." The figures have not been audited externally, and the extent to which they are reproducible in other organizational contexts remains an open question.
Prepared with AI assistance and reviewed by the editorial team.