Anthropic's latest disclosure shows that Claude now generates over 80% of the company's merged code in production environments, a significant increase from the single-digit percentage before its February 2025 release. This indicates that AI is becoming more deeply integrated into the actual software development process.
The code percentage has risen to over 80%.
The company disclosed that Claude has now become the primary code generation tool in its internal engineering processes. The production code referred to here is code that has been merged and deployed to the production environment, not draft code from the testing phase.
- Currently, over 80% of the code is in production.
- Before its release, the percentage was only in the single digits.
- The comparison point is around February 2025.
Capable of continuously processing long-term engineering tasks
Anthropic also stated that Claude can now complete software engineering tasks lasting up to 12 hours. These tasks typically involve more than just code completion; they may also involve debugging, modification, testing, and multi-step collaboration.
This means that the use cases for the model are expanding from short-term question-and-answer sessions and fragmented programming to more complete engineering execution processes.
The company discussed the future evolution of AI.
In this update, Anthropic also outlined several possible development scenarios, including a slowdown in AI progress and AI recursively self-improving, where AI systems help advance the development of the next generation of AI.
Supporters believe that such enhancements could accelerate progress in fields such as medicine, mathematics, and scientific research. However, critics argue that this is more like a continuous enhancement of programming-assisted tools and is insufficient to prove that general artificial intelligence is close to being realized.
Additional information:The data in this article comes from Anthropic's own disclosure and mainly reflects its usage in its internal development process, rather than being the result of independent third-party measurements.
