Anthropic Publishes First Concrete Evidence of 'Prosaic' Recursive Self-Improvement — 8x More Code Merged in 2026

The Anthropic Institute (via Import AI 460)·high signal

In a new Anthropic Institute post surfaced by Import AI 460, the company reports an 8x increase in lines of code merged into its codebase in 2026 versus 2021–2024, a trend that began in 2025 and 'accelerated significantly' as models got more capable. Jack Clark frames this as preliminary signs of the outer loop of RSI at a lab level — distinct from the maximalist version (AI designing its successor), which he estimates at 60% likely by end of 2028. For builders, it's the first first-party quantification of compounding AI-on-AI productivity inside a frontier lab.