The global artificial intelligence sector is reaching a financial boiling point as top laboratories approach trillion-dollar valuations and race toward autonomous engineering. On Thursday, Anthropic confirmed that its AI model, Claude, now authors more than 80% of the code merged into the company’s production codebase.
The operational shift is outlined in a newly published report titled “When AI Builds Itself.” Anthropic engineers are currently shipping roughly eight times as much code per quarter compared to their 2021 to 2025 baseline. Claude’s success rate on highly complex, open-ended coding problems jumped 50 percentage points over a six-month window, officially hitting 76% in May.
This rapid shift from manual coding to autonomous generation connects directly to incoming federal action. Two U.S. House lawmakers unveiled bipartisan AI legislation on Thursday designed to explicitly override state-level laws and mandate strict federal risk-management plans for top developers.
Inside Anthropic, the traditional software development pipeline is breaking down under the speed of the models. The company noted that humans are now the primary bottleneck in AI development. The daily workflow for staff is moving away from writing raw software and transitioning entirely into “architectural oversight” to review the output generated by AI agents.
How the Automated Factory Architecture Impacts the Industry
This data provides the first measurable evidence of recursive self-improvement inside a frontier AI laboratory. By offloading the vast majority of its engineering output to an autonomous model, Anthropic transitioned from building a helpful developer assistant to running a continuous automated factory.
The resulting acceleration aligns closely with urgent timelines published by competing laboratories. Google DeepMind CEO Demis Hassabis recently targeted Artificial General Intelligence for 2030 “plus or minus a year,” while Elon Musk predicts the threshold could be crossed by the end of this year. Anthropic is using its new productivity metrics to urge a global pause on frontier AI development, suggesting an international agreement with verification mechanisms to allow societal structures time to catch up.