Claude Code now runs daily maintenance on Anthropic’s software with a 46 percent merge rate

What changed

Anthropic began running daily maintenance on its own software using Claude Code, its in-house AI coding assistant. Over several weeks, Claude Code generated 388 pull requests targeting tasks like crash fuzzing and dead-code removal. After human review, 46 percent of those requests were merged into production. Claude Code’s inventor Boris Cherny describes this as early evidence that AI-assisted code maintenance at scale is achievable.

Why builders should care

Maintaining software requires constant, repetitive work that consumes developer time. Claude Code cutting through these routine tasks means teams can free up engineering resources for higher-value work. A 46 percent merge rate after human vetting shows AI can surface genuinely useful fixes without overwhelming reviewers with noise. This shifts how teams might integrate AI: not just for writing new features but automating ongoing upkeep.

The practical takeaway

Teams still need human oversight, but AI is proving capable enough to produce maintenance changes that meet real code standards. Builders should expect AI-assisted maintenance workflows to grow more common and necessary to keep codebases healthy while scaling fast. Early adopters who fine-tune review processes for AI-generated updates may see sustained productivity gains and fewer bugs slipping through.

What to watch next

Watch for how Claude Code and similar tools improve merge rates and reduce false positives. The speed and safety of continually relying on AI to manage production software maintenance will determine if this approach spreads beyond Anthropic. Industry watchers should monitor how feedback loops evolve between human engineers and AI to balance quality control with automation gains.

AI Quick Briefs Editorial Desk