Anthropic reports that its AI systems are now taking on more development tasks independently, speeding up AI progress significantly [1, 2]. The process, known as recursive self-improvement, involves AI building, testing, and enhancing itself without direct human coding input [1, 2].
Since 2021, Anthropic’s engineers have shipped eight times more code per quarter than in previous years, reflecting this shift toward AI-driven development [1]. From 2021 to 2023, the company built its first Claude chatbot using traditional human coding methods [1]. Between 2023 and 2025, chatbots were used to assist by generating small code snippets [1]. Moving into 2025 to 2026, Anthropic deployed autonomous coding agents that could write and edit entire files on their own [1]. By 2026, these agents gained the ability to run code themselves and delegate tasks to other agents [1].
Anthropic’s Jack Clark said, "In the near future, AI systems could become capable enough to autonomously design, build and train more capable successors on their own." He added that if this happens, "each new version of Claude could be built by the version before it, without human involvement" [2]. The company has observed that the length of AI tasks that can be reliably completed alone has been doubling roughly every four months recently [1].
Looking ahead, AI systems may fully close the loop by autonomously building and training their successors, though this remains a medium-confidence forecast [1, 2]. Jack Clark emphasized the need for societal tools "to validate and verify that the stuff being done by these AI systems is correct and is aligned with human intentions aligned with a thriving society" [2].
OpenAI has also publicly expressed concerns about risks related to recursive self-improvement, especially if researchers fail to share findings openly [2]. Anthropic plans to engage lawmakers soon to prepare them for the implications of this accelerating trend [2]. Clark said, "We've always found that the best thing to do is to socialize the concept and basically give people a sense of what's coming" [2].
In the coming months, Anthropic will begin discussions with legislators to address the challenges and policies needed for AI systems that can potentially build their own successors [2].
