Companies Discover AI Agent Costs Exceed Human Worker Salaries

Companies are learning a hard lesson: the cost of deploying AI agents to automate tasks can exceed the salaries of the human workers they replace. According to a report by Axios, the cumulative expense of AI agent requests—especially for high-volume tasks like code generation—often outstrips the wages of employees performing the same work.

AI agents are capable of handling a wide range of tasks, from repetitive duties to complex problem-solving. However, their most common workplace application is generating large volumes of code at speeds unattainable by human developers. In some cases, software engineers deploy multiple AI agents simultaneously, each working on separate tasks without direct supervision. These operations incur costs based on token usage, which can accumulate rapidly.

“For my team, the cost of compute is far beyond the costs of the employees,” said Bryan Catanzaro, vice president of applied deep learning at Nvidia, in an interview with Axios.

The issue has grown more pressing as organizations—including those developing AI tools—become increasingly dependent on AI automation. Anthropic now claims that “pretty much 100 percent” of its codebase is AI-generated, according to Boris Cherny, head of Claude Code, speaking earlier this year. Meanwhile, Google and Microsoft estimate that about a quarter of their code is AI-produced.

At Meta, employee performance reviews now include metrics on AI usage, signaling a top-down push toward automation. The trend has even spawned a competitive culture among tech workers, who are treating token consumption as a status symbol. The phenomenon, dubbed “tokenmaxxing,” sees some power users incurring monthly token bills exceeding $150,000.

“I probably spend more than my salary on Claude,” admitted Max Linder, a software engineer in Stockholm, in a recent interview with The New York Times. Meanwhile, engineers at Uber have reportedly exhausted the company’s entire 2026 AI budget using Claude Code, according to The Information.

Tech leaders are scrambling to address the issue, sometimes with unconventional solutions. In March, Nvidia CEO Jensen Huang proposed allocating AI tokens to software engineers equivalent to roughly half their base salary—a move framed as both a perk and a recruitment tool. Yet the approach underscores the financial strain on companies.

The situation also presents a lucrative opportunity for AI providers. One investor in OpenAI told Axios that the growing concern over token costs could benefit the company, citing OpenAI’s Codex as more token-efficient than Anthropic’s Claude Code. Anthropic has responded by raising its pricing, further increasing costs for heavy users.

The broader implications of AI automation remain uncertain. While some studies suggest efficiency gains, others highlight the risks of error-prone AI systems causing internal disruptions. Incidents at Meta and Amazon serve as cautionary examples. The debate continues over whether AI-driven automation is truly worth the potential costs and operational risks.