AI Is Reshaping Software Engineering: Le

AI Is Reshaping Software Engineering: Lead Developer of Claude Code Reveals 100% of Code Is AI-Generated

 

Amid the global wave of AI coding tools that are restructuring the industry's developmental logic, a set of data revealed recently by Boris Cherny, the lead developer of Anthropic's Claude Code team, has marked a groundbreaking milestone for the future of software engineering: 100% of the code he submits is generated by Claude Code, a status that has lasted for more than two months. This is not only the ultimate embodiment of AI coding tools' efficiency in individual work scenarios, but also signals a fundamental shift in the underlying logic of software development. The collaboration model between human engineers and AI is evolving from a "human-led, AI-assisted" paradigm to an entirely new one of "humans set the direction, AI executes the work".

 

The disruptive nature of Claude Code lies in its break from the limitations of traditional code completion tools, emerging as a truly autonomous, planning-capable agent-based AI coding assistant. Built on Anthropic's Claude 4 model series, it features an ultra-large 200k context window that enables seamless comprehension of complex large-scale codebase architectures. Natively supporting over 40 programming languages and mainstream development toolchains, it is also capable of independently completing multi-file modifications, debugging bugs, processing Pull Requests (PRs), and even resolving code merge conflicts, achieving full coverage of the development workflow from requirement implementation to code submission. It is for these reasons that the AI coding penetration rate within Anthropic has reached an unprecedented industry high: the code generation rate of the core team stands at approximately 90%, and across the entire company, it hits 70% to 90%. Behind these figures lies a core shift in the role of engineers—the once time-consuming and labor-intensive task of line-by-line coding is gradually being taken over by AI, and human engineers have refocused their work on the more valuable aspects of accurately defining business requirements, designing sound system architectures, and conducting final code quality reviews.

 

The strength of Claude Code is most powerfully demonstrated by the team's deep practice of "eating their own dog food". As the creator of Claude Code, Boris Cherny is not only its ultimate practitioner; the team he leads has also fully implemented the AI collaborative development model: 3 to 8 Claude instances are run in parallel for daily development, and all code undergoes dual reviews by both humans and AI, ensuring both development efficiency and uncompromising code quality. More remarkably, the core code of Claude Code (Claude Cowork) was entirely written by AI in just 1.5 weeks. This achievement not only fully validates the enormous potential of AI in complex engineering development, but also turns the concept of "developing AI tools with AI" into a reality, serving as the perfect testament to the capabilities of AI coding.

 

Today, the changes brought about by Claude Code have long transcended the efficiency gains of a single tool, offering a brand-new developmental insight for the entire software engineering industry. Boris Cherny revealed that he submits 22 to 27 PRs a day, all independently completed by Claude. Such efficiency improvements are restructuring the human collaboration models and value distribution logic of software engineering. In the future, the core competitiveness of software development will no longer lie in engineers' coding speed and proficiency, but in their ability to understand business requirements, design system architectures, and collaborate with AI. The practice of Claude Code has also made it clear to the industry that AI is not a replacement for human engineers, but a powerful collaborative partner that liberates them from repetitive work while driving the entire software engineering industry toward greater efficiency and higher value.