Humans say AI is already developing AI, and humans may be the ones slowing things down

Artificial intelligence has become so efficient at writing code and research that the biggest barrier to developing new AI systems may now be the humans who supervise them, according to a new study by Anthropic.

In her report “When Artificial Intelligence Builds Itself” published On Thursday, Anthropic claimed that Cloud is already helping build future AI systems by writing code, running experiments, and assisting with research — a trend the company says could eventually lead to iterative self-improvement, as AI systems help design their successors.

Claude now authored more than 80% of the code built into its code base, and has helped engineers increase code production nearly eight-fold since 2024, Anthropic said.

“Prior to Cloudcode’s launch in Search Preview in February 2025, this number was in the low single digits,” Anthropic wrote, adding that the shift is also evident in the amount of production per engineer. “The lines of code merged per engineer daily remained flat for Anthropic’s first four years (2021-2024), then started to climb in 2025 when Claude started running the code instead of just suggesting they copy and paste it to the engineer.”

Anthropic said the future could unfold in several ways: AI progress could slow, humans could remain in charge while AI automates much of the work, or AI systems could eventually begin to improve on their successors.

“Given enough time, and with enough computing power, this trend points to an AI system capable of designing and developing its successor completely autonomously,” Anthropic wrote. “This is called iterative self-improvement. We are not there yet, and iterative self-improvement is not inevitable. But it may come sooner than most organizations are preparing for it.”

The company said it’s too early to know what outcome is most likely, but it argues that AI is already helping build AI, and acknowledged that lines of code are an imperfect measure of productivity.

“None of this is a guarantee that iterative self-improvement is on the horizon,” Anthropy later said books On

The report comes as AI companies increasingly position their models as research collaborators rather than simple chatbots. However, Anthropic said the increase in code production reflects a broader acceleration in software development driven by increasingly capable AI agents.

Last month, Anthropic updated its flagship Claude model to Opus 4.8a continuation of a steady stream of releases aimed at improving coding, reasoning, and performance of autonomous tasks. Meanwhile, rival developer OpenAI followed a similar strategy with the launch of its frontier models GPT-5.5 and GPT-Rosalind In April.

In May, Google announced Gemini SparkA personal AI agent that doesn’t wait to be asked. It manages tasks across apps, flags items that need attention, and finishes tasks in the background.

The report also comes as Anthropic has increasingly emphasized AI systems capable of operating more autonomously as it prepares to go public. In recent months, Anthropic has reviewed developments in programming, agent workflows, and long-running task performance, while promoting Claude MythosAbility to identify software vulnerabilities and conduct complex cybersecurity research.

“Humans play a greatly diminished role in their own development, and we will likely shift most of our efforts toward monitoring, verifying, and validating an expanded ‘virtual laboratory’ run by AI systems,” the company said. “We expect that systems capable of automated AI R&D will have skills that will transfer to the rest of science, allowing them to begin to revolutionize other fields.”