Summary
- Anthropic reports that Claude generates over 80% of the code integrated into its codebase.
- The AI firm claims its engineers are producing approximately eight times more code than they did in 2024.
- Anthropic suggests that AI is already contributing to the development of future AI systems and may eventually aid in creating its own successors.
According to a recent analysis by Anthropic, the rapid advancement of AI in coding and research suggests that the primary hindrance to the evolution of new AI systems may now be the human supervisors managing them.
In a report titled “When AI Builds Itself,” which was released on Thursday, Anthropic asserted that Claude is actively participating in the development of future AI systems by writing code, conducting experiments, and supporting research efforts. The company believes this trend could lead to recursive self-improvement, where AI systems contribute to designing their own successors.
The report states that Claude is responsible for more than 80% of the code that has been added to its codebase, and since 2024, it has assisted engineers in increasing their code production by approximately eight times.
“Prior to the launch of Claude Code in research preview in February 2025, this figure was in the low single digits,” Anthropic noted, highlighting a significant increase in output per engineer. “The lines of code merged per engineer daily remained stable during Anthropic’s initial four years (2021-2024), but began to rise in 2025 when Claude started executing code rather than merely suggesting it for engineers to implement.”
Anthropic outlined several potential scenarios for the future: progress in AI could decelerate, humans could continue to oversee while AI automates many tasks, or AI systems might start enhancing their own successors.
“If this trend continues and sufficient computational power is available, it could lead to an AI system capable of fully autonomously designing and developing its own successor,” Anthropic explained. “This concept is known as recursive self-improvement. While we are not at that point yet, and its inevitability is uncertain, it may occur sooner than many organizations anticipate.”
The company emphasized that it’s premature to predict which outcome is most plausible, yet they maintain that AI is already aiding in AI development, acknowledging that measuring productivity solely by lines of code has its limitations.
There are no guarantees that recursive self-improvement is imminent. It remains unclear whether Claude possesses the research judgment necessary to identify the most appropriate problems to tackle.
However, if these trends persist, the idea of AI systems designing and constructing their own successors becomes conceivable. This…
— Anthropic (@AnthropicAI) June 4, 2026
“There are no guarantees that recursive self-improvement is imminent,” Anthropic reiterated on X. “It’s still uncertain whether Claude can exercise research judgment—specifically, selecting the right problems to address.”
This report arrives as AI companies increasingly present their models as collaborative research partners rather than mere chatbots. Nevertheless, Anthropic noted that the rise in code production mirrors a broader acceleration in software development, driven by increasingly advanced AI agents.
Last month, Anthropic upgraded its flagship model, Claude, to Opus 4.8, continuing its trend of regular updates aimed at enhancing coding capabilities, reasoning, and autonomous task execution. In parallel, competitor OpenAI has followed a similar approach with its latest models, launching GPT-5.5 and GPT-Rosalind in April.
In May, Google introduced Gemini Spark, an AI agent that autonomously manages tasks across applications, flags items that require attention, and completes jobs in the background.
This report also comes as Anthropic has increasingly focused on AI systems capable of greater autonomy in anticipation of going public. Recently, the company has highlighted advancements in coding, agentic workflows, and long-duration task capabilities, while promoting Claude Mythos’s proficiency in identifying software vulnerabilities and conducting complex cybersecurity research.
“Humans are likely to play a significantly reduced role in the development process, shifting most of our efforts toward overseeing, validating, and verifying an expanding 'virtual lab' operated by AI systems,” the company stated. “We anticipate that systems capable of automated AI research and development will acquire skills that can be applied to other scientific fields, potentially revolutionizing those areas as well.”