At Anthropic—an AI lab developing some of the world’s most advanced models—engineers no longer write the code powering their products; they’re letting AI handle it. The has stated he hasn’t written any code in more than two months.
, Cherny noted that 100% of his code is now produced by Anthropic’s Claude Code and Opus 4.5. Across the rest of the company, he claims “nearly 100%” of code is also AI-generated.
“For me personally, it’s been 100% for over two months now—I don’t even make minor manual edits,” Cherny wrote in an X post responding to AI researcher Andrej Karpathy. “I merged 22 pull requests (PRs) yesterday and 27 the day prior, each entirely written by Claude.”
These comments mirror earlier statements from Anthropic CEO Dario Amodei at the World Economic Forum this month, where he noted that some engineers at his firm have stopped writing code themselves and instead rely on AI models to generate it while they focus on editing. At Davos, Amodei predicted the industry could be just six to twelve months away from AI handling most or all of software engineering work from start to finish.
Cherny isn’t the only prominent engineer to announce they’ve largely stopped manual coding. Roon—a popular pseudonymous X account run by an OpenAI researcher—also stated in a post that he no longer writes any of his own code. “100%, I don’t write code anymore,” the responded when asked what share of his coding is done by AI models. In another post, he added: “Programming always was a hassle. It was a necessary pain for almost everyone wanting to get computers to do useful things, and I’m relieved it’s over.”
While those in the industry do have reasons to promote their own tools, there’s a growing agreement that the industry has already been fundamentally transformed by the rise of AI coding tools.
“The entire way people build software has changed; software isn’t what it used to be,” Himanshu Tyagi, co-founder of open-source AGI firm Sentient, told . “A large portion of the code shipped over the next 10 years will be written by AI…[Anthropic’s] Claude Code is the breakthrough product that made this possible.”
Still, outside leading AI labs, the AI-generated code figures reported by many software companies are much lower. For example, CEO Satya Nadella said in April 2025 that AI generates around 30% of code at the tech giant. has shared a similar number. A study published in the journal Science earlier this month analyzed GitHub Python functions and found that about 29% in the U.S. are now AI-written, with lower rates in other regions.
And while Cherny claims AI now writes 100% of his code, an Anthropic spokesperson stated the company-wide figure is between 70% and 90%. For Claude Code itself, roughly 90% of its code is generated by Claude Code.
Cherny, for his part, believes these numbers will keep rising and that other companies will soon reach similar AI code generation rates. “I think most of the industry will see similar stats in the coming months—it’ll take longer for some than others,” he wrote. “We’ll then start seeing comparable figures for non-coding computer tasks too.”
Anthropic’s tools have grown popular among software engineers over the past few years. But the release of Claude Code struck a chord with both coders and non-coders, sparking a viral moment for the company not seen since ChatGPT’s launch. After users pointed out Claude Code was more of a general-purpose AI agent, Anthropic created a non-coder version: Cowork, a file management agent that’s essentially a user-friendly iteration of the coding tool. Cherny said his team built Cowork in about a week and a half, mostly using Claude Code itself.
Even before the public buzz, Cherny says the tool was gaining traction within his company.
“Around a year ago…we had the idea that the model was powerful enough to use for a different type of coding…we started testing it internally, and it immediately took off,” Cherny told in an interview last week. “I’ve never enjoyed my day-to-day work this much right now—Claude handles almost all the tedious tasks, so I get to be creative and think about what to build next.”
Cherny also uses Claude Code for various administrative tasks outside coding, including project management work like automatically messaging team members on Slack when they haven’t updated shared spreadsheets.
“Engineers just feel freed—they don’t have to work on all the tedious stuff anymore,” he said.
A reckoning for the software industry
The rise of AI-generated code has significantly affected the software industry. Many Big Tech companies have been transparent about AI models writing large amounts of their code. But automating much of the coding process has also raised questions about the future of software engineering roles, especially entry-level positions that traditionally served as training grounds for the field.
Tech companies argue that rapid adoption of AI coding tools like Claude Code and GitHub Copilot will democratize coding, letting people with little to no technical skills build products by prompting AI systems in natural language. However, while it’s not confirmed that the two are causally linked—and other factors impact job declines—entry-level software engineer openings have indeed dropped as generative AI code output has risen.
The shift is already altering how Anthropic approaches hiring. Cherny said his team now mostly hires generalists instead of specialists, as many traditional programming skills are less relevant when AI handles implementation details.
“Not all the skills people learned in the past apply to coding with LLMs,” Cherny wrote. “The model can fill in the details.”
While Cherny highlighted the productivity boosts and creative freedom AI coding tools offer, he also acknowledged the technology is still evolving. According to Karpathy’s analysis, models can make “subtle conceptual errors,” overcomplicate code, and leave dead code behind. Despite these limitations, engineers like Cherny are confident AI-generated code quality will only keep improving.