Engineers at leading artificial intelligence developer Anthropic have announced that nearly all code powering their products is now generated by AI, specifically Claude Code and Opus 4.5. Boris Cherny, head of Anthropic’s Claude Code, stated on X that he has not written code manually for over two months, shipping dozens of pull requests daily that were entirely AI-written.
These reports echo sentiments from within OpenAI, where a pseudonymous researcher known as Roon also confirmed that one hundred percent of their coding tasks are now handled by AI models. Roon expressed relief, noting that programming was a "requisite pain" that the industry is now moving beyond, according to posts on X.
Anthropic CEO Dario Amodei previously projected at the World Economic Forum that the industry might see AI handling the majority of software engineering tasks within six to twelve months. While industry insiders may have incentives to promote their tools, Himanshu Tyagi, co-founder of Sentient, confirmed that Claude Code represents a breakthrough product fundamentally changing software construction.
However, this level of automation remains largely confined to the forefront AI labs, as reported by Fortune. Microsoft CEO Satya Nadella indicated that AI generated approximately 30% of code at the software giant last year, a figure consistent with other major firms like Salesforce.
Anthropic itself reports a company-wide code generation rate between 70% and 90%, though Cherny maintains that individual productivity is reaching the one hundred percent mark. Cherny anticipates that other companies will reach similar statistics in the coming months, eventually extending to non-coding computer tasks as well.
This rapid adoption is already influencing hiring strategies at Anthropic, where teams now prioritize generalists over specialists, as AI handles complex implementation details. Cherny noted that traditional programming skills are becoming less relevant when large language models can fill in the necessary specifics.
Despite the productivity gains Cherny experiences, he acknowledged that models can still produce subtle conceptual errors or overly complex solutions. Nevertheless, the confidence among these leading engineers is high that the quality of AI-generated code will continue its upward trajectory.
