There had been some eyebrows raised when Dario Amodei had predicted that AI would write 90% of the code in 3-6 months, but he says that his prediction has come true at his own company.
In a conversation with Salesforce CEO Marc Benioff, Anthropic’s CEO Dario Amodei revealed that his prediction about AI-generated code is now a reality within his own organization. The revelation offers a rare glimpse into how one of the world’s leading AI companies is using its own technology to accelerate development and solve complex engineering challenges.
Amodei acknowledged the skepticism his prediction had generated, but stood by his claim. “Six months ago, I made this prediction that in six months, 90% of code would be written by AI models. Some people think that prediction is wrong. But within Anthropic and within a number of companies that we work with, that is absolutely true,” he said.
When Benioff pressed him on the specifics—asking if 90% of all code at Anthropic was being written by Claude today—Amodei clarified that while the reality varies across teams, the transformation is substantial. “Not uniformly,” he admitted, but emphasized that the shift has fundamentally changed how engineering teams operate. “The teams play an editing and supervisory role. But Claude can—you can say write this feature and Claude will do it, and the human kind of looks over the code or asks for help from another model, another copy of Claude.”
The Anthropic CEO went further, sharing a striking example of Claude’s debugging capabilities. “Sometimes we have these long-running bugs. When we were training the most recently released Claude, the cluster was broken. There was this bug. Engineers were spending a few days on it, and we just went to Claude and we said, ‘I don’t know, just kind of play around with the cluster and see if you can figure out what’s wrong.’ And Claude found this just super obscure bug that the engineers had missed,” Amodei recounted. “This was one area where it was like Claude is like having a very talented teammate.”
Perhaps most tellingly, Amodei revealed that Anthropic is using Claude to build upon itself: “Within Anthropic and elsewhere, we are using Claude to help build products on top of Claude and to help train the next Claude, to help serve Claude faster.” This recursive improvement loop—AI helping to build better AI—represents a significant milestone in the technology’s development.
The implications of Amodei’s comments extend far beyond Anthropic’s offices. If AI can truly handle 90% of coding tasks at one of the world’s most sophisticated AI companies, it suggests a fundamental shift in software development is underway. This aligns with recent trends across the tech industry, where AI coding assistants have moved from experimental tools to essential parts of developers’ workflows. GitHub reported that developers using its Copilot tool accept AI suggestions for nearly 30% of their code, and that figure continues to climb. Major tech companies including Google, Meta, and Microsoft have similarly integrated AI coding tools into their development processes. However, Amodei’s claim represents a big leap beyond these incremental adoptions—suggesting that the role of human developers is evolving from writers to editors and supervisors, fundamentally redefining what it means to build software in the age of AI.