Claude Writes 80% of Its Own Code: Anthropic Reaches Milestone in AI Self-Development

🔄 Update — June 06, 2026: Recursive Self-Improvement and Safety Warning

Anthropic has published a detailed report on “recursive self-improvement,” warning of a rapid loss of human control. Claude now writes over 80% of production code—a massive jump from less than 10% in early 2025.

What’s new?

Recursive Self-Improvement: Anthropic warns that AI systems could optimize themselves faster than anticipated, leading to oversight challenges.
Production Code Milestone: More than 80% of new code at Anthropic is generated by Claude, highlighting the unprecedented speed of development.
Foundry Managed Compute: In parallel, Microsoft’s new Foundry Managed Compute provides the infrastructure layer required for the persistent operation of such autonomous agent workloads.
Call for a Pause: Due to these risks, Anthropic is urging a temporary pause on frontier AI development to discuss safety protocols.

Why this adds to the article

This update clarifies that the previously reported 80% milestone is not just a productivity win but carries serious existential safety risks due to uncontrolled self-optimization.

Summary

Anthropic has announced that its Claude AI system now writes and merges over 80% of its own codebase. This breakthrough marks a significant shift toward autonomous AI self-development and highlights the maturation of agentic coding workflows in real-world applications.

What happened?

In recent reports and social media statements, Anthropic executives revealed that Claude has become the primary developer of its own successor. 80% of the code merged into their repository is AI-generated. This comes alongside the launch of Claude Opus 4.8 and Anthropic’s confidential IPO filing, showcasing the company’s massive technical and commercial momentum.

Why it matters

This milestone signals the arrival of a “self-replication” phase, where AI massively accelerates its own improvement. For software engineering, this represents a paradigm shift: the role of the human developer is evolving from a code writer to a code reviewer and architect.

Evidence

The information is based on reports from Tom’s Hardware and confirmations across social channels like LinkedIn and Instagram. Furthermore, Anthropic has confidentially filed an IPO prospectus with the SEC, reflecting confidence in this technological leadership.

Analysis

The high rate of 80% merged code suggests that the AI is not just handling trivial tasks but is deeply integrated into the core architecture. While the exact definition of “merged code” (lines vs. modules) is still being discussed, the trend toward full autonomy is unmistakable.

Practical Takeaways

Agentic Workflows: Organizations should begin integrating agentic workflows into their development cycles.
Skill Shift: Focus on system design and validation rather than just syntax proficiency.
Acceleration: AI model development cycles will continue to shorten due to self-optimization.

Open Questions

What will quality assurance look like with 100% autonomous code?
What are the implications for the long-term maintainability of systems that humans may no longer fully understand?