The AI That Now Writes Most of Its Maker's Code

Short summary

Anthropic reports 80% of shipped code is now written by Claude, shifting engineers from authoring to reviewing. More significantly, internal models outperform human researchers at choosing research directions and autonomous agents solved near-intractable AI safety problems with minimal oversight. These breakthrough findings come from unreleased proprietary models requiring external verification.

• •80% of Anthropic's shipped code now written by Claude, up from single digits two years ago
• •AI models now choose research directions better than human scientists and solve complex problems autonomously
• •Findings are from unreleased internal models and require external verification before broader acceptance

Generated with AI, which can make mistakes.