Anthropic's most frequent user of its own Claude AI isn't an engineer — it's the head of tax. Chief financial officer Krishna Rao disclosed the fact during a recent internal presentation, offering a rare look at how the company's finance team has become a surprising powerhouse of AI adoption.
The Tax Chief's Unnoticed Record
Rao said the head of tax consumes more Claude tokens than anyone else at Anthropic. The revelation underscores a pattern the finance team has observed: the company's heaviest AI users are accountants, not the software engineers building the models. The tax chief's daily workflow, which involves parsing complex tax codes and reviewing contract language, turned out to be a perfect fit for Claude's capabilities.
Finance Team's Custom Toolkit
The finance group hasn't just been passive consumers. Over time, they've built up to 150 custom AI tools tailored to their specific tasks. These range from automated expense report auditors to contract clause checkers. Rao noted that the team's enthusiasm for building and deploying these tools has made them a sort of internal test bed for real-world corporate AI use cases — far removed from the typical demo scenarios of code generation or creative writing.
Why Accountants Lead the AI Usage
The pattern bucks the common assumption that technical staff would be the most aggressive adopters of a company's own AI. But Rao's disclosure suggests that the real value of large language models in a corporate setting may lie in rote, high-volume document work — the kind that dominates a finance department's day. Anthropic's accountants, he explained, saw early on that Claude could cut hours of manual review into minutes. The result: a quiet revolution in the back office, not the engineering floor.
Rao didn't say exactly how many tokens the tax head uses per month, but the fact that a single finance executive tops the company-wide leaderboard is telling. It raises a question that other companies building AI tools are now likely asking themselves: are we looking for power users in the wrong departments?
