When AI Writes and Reviews the Code (r.rich)

Anthropic recently revealed that its AI model, Claude, is now responsible for writing over 80% of the code integrated into its production codebase, a dramatic increase from the low single digits since its launch in February 2025. As AI increasingly takes on both coding and review roles, this shift raises critical questions about the effectiveness and authenticity of traditional code review mechanisms, historically reliant on human oversight. What was once an attentive audit is now becoming a mere formality, where human reviewers often only rubber-stamp changes made by AI, leading to potential auditing discrepancies. The implications for the AI and ML community are significant, as this evolution prompts a rethinking of code governance and quality assurance. Instead of focusing solely on peer code review, Anthropic suggests transitioning to an "AI-Native Change Management" (ANCM) model, emphasizing independent verification through automated testing, adherence to defined specifications, and segregating duties between writing and reviewing agents. By utilizing automated tools to scan for bugs and implementing structured verification plans, humans can concentrate on higher-level decision-making regarding the objectives of the code changes. This approach maintains accountability and reduces biases typically present in a unified reviewer-writer model, fostering a more reliable and effective coding environment in the era of AI-driven development.