•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
Fortune reports that at leading AI labs Anthropic and OpenAI, software developers are increasingly handing over coding tasks to AI systems—shifting from writing code themselves to overseeing and refining AI-generated output.
At Anthropic, Boris Cherny, head of the Claude Code project, said he has not written any code manually for more than two months. On X, Cherny claimed that 100% of the source code he uses is produced by Claude Code and Opus 4.5, and that the same approach applies to much of his team.
In comments to AI researcher Andrej Karpathy, Cherny said he has maintained 100% AI-generated code for more than two months and no longer performs manual edits. He also cited AI-driven output volumes, including up to 22 pull requests in a day and 27 the previous day, which he said were written by AI.
At OpenAI, Roon, a researcher, similarly said he has stopped writing code manually. When asked about the share of AI in his work, Roon replied: “100%, I don’t write code anymore.”
Anthropic CEO Dario Amodei, speaking at the World Economic Forum in Davos earlier this month, said the industry is six to twelve months away from AI handling the full software engineering process from start to finish. He said that, for now, engineers have shifted toward editing and quality control rather than direct coding.
Fortune points to the Cowork project, described as a file-management tool for non-specialists, as an example of the pace of change. Cherny’s team reportedly took about a week and a half to build the product, largely aided by Claude Code.
The report also says AI is being used for administrative work, including project management and automated reminders on Slack when colleagues forget to update a shared spreadsheet.
Despite rapid progress, the report highlights gaps across the broader software industry. In data published in April 2025, Microsoft CEO Satya Nadella said AI contributes about 30% of code within the company. Salesforce reported similar figures.
A Science journal study earlier this month found that about 29% of Python functions on GitHub in the US are AI-written, with lower rates outside the US. At Anthropic, however, the company-wide average is reported to range from 70% to 90%. For Claude Code specifically, around 90% of its source code is written by the system itself.
The rise of AI coding is raising questions about the future of programming—particularly for entry-level roles that have traditionally served as a training ground for new engineers. Fortune says Anthropic has already adjusted recruitment, prioritizing candidates with broad thinking rather than specialists in a specific language or technology.
Cherny emphasized that not all past learning transfers directly to programming with large language models, adding that AI can fill in details while humans provide creativity.
Even with increased automation, Karpathy cautioned that AI models can still make subtle conceptual mistakes. These errors can complicate code or leave behind unnecessary “garbage” code, requiring human oversight.
For Cherny, the shift away from repetitive tasks is described as a major personal benefit. He said: “I have never enjoyed my work as much as I do now. Claude handles the dull parts, and I am free to create and think about what I want to build next.”
Premium gym chains are entering a “golden era” that is ending or already in decline, as rising operating costs collide with shifting consumer preferences toward more flexible, community-based ways to exercise. Long-term memberships are shrinking, margins are pressured by higher rents and facility expenses, and competition from smaller, more personalized…
