Anthropic has announced that its Claude model is currently involved in the development of its own successor. According to the company, the AI is independently handling 26% of all research and development tasks for the upcoming version, operating from start to finish based on high-level prompts provided by human engineers.
In addition to independent tasks, Claude contributes to approximately 90% of implementation work through collaboration with human teams. While the model executes significant portions of the development process, the company clarifies that there is no fully autonomous mode for the AI, and all actions remain under human supervision.
The company's internal development workflow has evolved significantly since 2021. It transitioned from manual coding in 2021–2023 to short-form code generation in 2023–2025, and now utilizes agents capable of editing files and delegating tasks. This shift has allowed Anthropic engineers to increase their quarterly code output by 8 times.
Anthropic notes that while recursive self-improvement is not yet a reality, it remains a possibility with increased computing power and improved AI capabilities. Consequently, the company is prioritizing enhanced safety protocols, system monitoring, and behavioral control measures to manage the risks associated with these advanced development systems.
