Anthropic’s chief financial officer, Krishna Rao, said in a podcast interview published May 13, 2026, that the company’s Claude models now write more than 90 percent of Anthropic’s own code and prepare most of its internal financial reporting before any person reviews it.
Speaking on the “Invest Like the Best” podcast, Rao said Claude plays a substantial role in building Anthropic’s own developer tools, including Claude Code, which means the model is contributing to the software used to develop its successors. Anthropic executives had previously said AI wrote a large majority of the company’s code; Rao’s comments put a figure on it from the finance side of the business.
Finance work handled by agents
Rao said the company’s monthly financial reports are now close to complete before humans look at them, with 90 to 95 percent of the analysis already produced by AI. He described what he called an army of agents working continuously across different projects in the finance organization, with staff shifting toward reviewing, checking and deciding rather than assembling the numbers.
The comments go beyond automated reporting. They describe a company reorganizing core corporate functions around AI systems while it sells those systems to other businesses, which makes Anthropic an early test of whether the productivity gains it promises customers show up in its own operations. Rao also said that Anthropic has continued to hire even as AI handles more of the work, according to coverage of the interview.
Revenue growth
The financial figures Rao discussed drew as much attention as the automation claims. Anthropic’s annualized revenue run rate was around $250 million roughly two years earlier. By the end of the first quarter of 2026, he said, the run rate had passed $30 billion. A run rate projects a recent period’s revenue over a full year and is not the same as reported annual revenue, but the trajectory is far steeper than that of traditional software companies at a comparable stage.
Rao also said most of Anthropic’s computing capacity is devoted to research and model training rather than to serving customers, and he discussed the company’s multiyear compute commitments, which have been reported to exceed $100 billion. Together, the comments sketch a business in which AI is both the product and, increasingly, the workforce.
