Sign up to see the future, today<\/p>\n
Can\u2019t-miss innovations from the bleeding edge of science and tech<\/p>\n
Mindlessly unleashing AI agents to take over employees\u2019 jobs can be pretty costly, it turns out. Some companies are learning the hard way that paying for the incredible volume of AI agent requests is costing more than what they\u2019d pay their human employees, Axios reports<\/a>.<\/p>\n AIs can perform all sorts of tasks, ranging from the rote to the complex. But one of the most popular ways it\u2019s being used in the workplace is to generate mountains of code at a pace far greater than a human could achieve. Sometimes, software engineers will even run multiple AI agents at the same time, all working on different tasks in the background without supervision. Each of these tasks costs tokens, and the bill can quickly add up.<\/p>\n \u201cFor my team, the cost of compute is far beyond the costs of the employees,\u201d Bryan Catanzaro, vice president of applied deep learning at Nvidia, told Axios.<\/p>\n The problem has become harder to ignore as organizations are increasingly reliant on using AI tools and agents \u2014 including the organizations building them.\u00a0\u201cPretty much 100 percent\u201d of Anthropic\u2019s code is now AI-generated, the company\u2019s head of Claude Code Boris Cherny claimed earlier this year<\/a>. Google and Microsoft\u2019s bosses claim that this share is around a quarter of their companies\u2019 code. Meta employees performance reviews are now partly based on how much AI they use<\/a>, showing that a lot of the push towards using AI is coming from the top.<\/p>\n It probably doesn\u2019t help that many tech workers are treating their token bills as member-measuring contests<\/a>, using millions of tokens in a single day. The slang for this, we regret to inform you, is \u201ctokenmaxxing,\u201d with some power users racking up monthly token bills north of $150,000. \u201cI probably spend more than my salary on Claude,\u201d Max Linder, a software engineer in Stockholm, told The New York Times last month<\/a>. Uber engineers using Claude Code have already blown through the company\u2019s entire 2026 AI budget, The Information reported<\/a>.<\/p>\n Tech leaders\u2019 attempts to grapple with the situation can sound nearly as comical as the dilemma itself<\/a>. In March, Nvidia CEO Jensen Huang proposed giving software engineers AI tokens equal to roughly half their base salary, something he said could be used as a recruiting tool. Why be wooed by a signing bonus, when if you work for us, you get to use more AI?<\/p>\n At the same time, it\u2019s a clear money-making opportunity for AI providers. One OpenAI investor told Axios that the concern over token costs could benefit them, since they believe Codex uses tokens more efficiently than Anthropic\u2019s Claude Code. Anthropic, meanwhile, has cashed in by raising its pricing.<\/p>\n In all, the token costs are just one of many major question marks over AI automation. The jury\u2019s still out on whether using error-prone AIs is more efficient and worth the potential havoc they can wreak internally \u2014 as evidenced by incidents at Meta<\/a> and Amazon<\/a>, among others \u2014 while numerous studies suggest<\/a> that forcing workers to use AI tools could actually be making their jobs harder.<\/p>\n