Claude Wrote 80% of Anthropic Code as CI Tasks Surge 25-Fold

Internal metrics from Anthropic reveal a striking reality: Claude now generates 80% of the company’s code, with engineers delivering eight times the quarterly output seen between 2021 and 2025. Even critical workflows like code reviews and merge approvals are increasingly automated by the AI.
Yet, AI’s coding behavior diverges sharply from human patterns. It favors submitting pull requests with extreme granularity, issuing continuous commits throughout the day, and triggering a tenfold surge in test cases. This relentless pace overwhelmed the CI system, causing task volumes to spike 25 times over six months. Despite three attempts to patch the system—adding core logic, sharding, and daily forced restarts—all efforts failed.
Ultimately, Anthropic followed Claude’s recommendation to rebuild the service entirely, adopting a new distributed, stateless architecture that resolved the CI overload. This incident highlights a crucial shift: AI’s impact has moved beyond whether programmers can write code to the capacity of the entire software engineering system. When coding is no longer bound by human rhythms, pipelines designed for human habits are often the first to break.
Related article
Google Renames NotebookLM to Gemini, Equipping It With a Cloud Computer Capable of Writing and Running Code
Longtime NotebookLM users should prepare for a familiar face with a new identity. On the 16th, Google announced that its AI note-taking assistant, NotebookLM, has been officially rebranded as Gemini Notebook, marking its full integration into Google’
Microsoft Launches New Venture, Investing $2.5 Billion to Deploy 6,000 Engineers at the Forefront of Enterprise Innovation
As artificial intelligence shifts from "model development" to "business implementation," Microsoft has once again demonstrated its strong resource integration capabilities. Recently, Microsoft announced the establishment of a new business unit called
Anthropic Hit by Two Major Data Leaks Within a Week, Shaking Trust in Its AI Safety Claims
Anthropic, the AI leader recognized for its emphasis on safety and caution, has recently caused a stir in the tech sector. Following an accidental release of 3,000 internal documents last week, the company committed another significant error this Tue
Related Special Topic Recommendations
Comments (0)
0/500

Internal metrics from Anthropic reveal a striking reality: Claude now generates 80% of the company’s code, with engineers delivering eight times the quarterly output seen between 2021 and 2025. Even critical workflows like code reviews and merge approvals are increasingly automated by the AI.
Yet, AI’s coding behavior diverges sharply from human patterns. It favors submitting pull requests with extreme granularity, issuing continuous commits throughout the day, and triggering a tenfold surge in test cases. This relentless pace overwhelmed the CI system, causing task volumes to spike 25 times over six months. Despite three attempts to patch the system—adding core logic, sharding, and daily forced restarts—all efforts failed.
Ultimately, Anthropic followed Claude’s recommendation to rebuild the service entirely, adopting a new distributed, stateless architecture that resolved the CI overload. This incident highlights a crucial shift: AI’s impact has moved beyond whether programmers can write code to the capacity of the entire software engineering system. When coding is no longer bound by human rhythms, pipelines designed for human habits are often the first to break.
Google Renames NotebookLM to Gemini, Equipping It With a Cloud Computer Capable of Writing and Running Code
Longtime NotebookLM users should prepare for a familiar face with a new identity. On the 16th, Google announced that its AI note-taking assistant, NotebookLM, has been officially rebranded as Gemini Notebook, marking its full integration into Google’
Microsoft Launches New Venture, Investing $2.5 Billion to Deploy 6,000 Engineers at the Forefront of Enterprise Innovation
As artificial intelligence shifts from "model development" to "business implementation," Microsoft has once again demonstrated its strong resource integration capabilities. Recently, Microsoft announced the establishment of a new business unit called
Anthropic Hit by Two Major Data Leaks Within a Week, Shaking Trust in Its AI Safety Claims
Anthropic, the AI leader recognized for its emphasis on safety and caution, has recently caused a stir in the tech sector. Following an accidental release of 3,000 internal documents last week, the company committed another significant error this Tue





Home






