It turns out that unleashing mindless AI agents to take over employees’ jobs can be quite costly. Some companies are learning the hard way that paying for the incredible volume of AI agent requests is more expensive than paying their human employees, axios reports.
AI can perform all types of tasks, from rote to complex. But one of the most popular methods used in the workplace is to generate mountains of code at speeds far beyond what a human can achieve. Sometimes, software engineers will even run multiple AI agents at the same time, working on different tasks in the background without supervision. Each of these tasks costs tokens, and the bill can add up quickly.
“For my team, the cost of compute far outweighs the cost of staff,” said Brian Catanzaro, vice president of applied deep learning at Nvidia. axios.
The problem has become harder to ignore as organizations become increasingly dependent on the use of AI tools and agents – including the organizations that create them. “Almost 100 percent” of Anthropic’s code is now AI-generated, Boris Cherny, the company’s head of cloud code, claimed earlier this year. The owners of Google and Microsoft claim this share is about a quarter of their companies’ code. Meta employees’ performance reviews are now based partly on how much AI they use, which suggests that a lot of pressure toward using AI is coming from the top.
This probably doesn’t help many technical staff Treating your token bills as a member-measuring competitionUsing millions of tokens in a single day. We regret to inform you that the slang for this is “tokenmaxing”, with some power users racking up monthly token bills in excess of $150,000. “I probably spend more than my salary on the cloud,” says Max Linder, a software engineer in Stockholm. told the new York Times last month. Uber engineers using cloud code have already blown the company’s entire 2026 AI budget, Information informed.
Tech leaders’ attempts to deal with the situation may seem poor This dilemma is as funny as it gets. In March, Nvidia CEO Jensen Huang proposed giving software engineers AI tokens equivalent to about half their base salary, saying it could be used as a recruiting tool. Why be tempted by a signing bonus when if you work for us, you get to use more AI?
Plus, this is clearly a money-making opportunity for AI providers. An OpenAI investor told axios Concerns over token costs may benefit him, as he believes Codex uses tokens more efficiently than Anthropic’s cloud code. Meanwhile, Anthropic has cashed in by raising its price.
Overall, token cost is one of the many major question marks over AI automation. The jury is still out on whether using error-prone AI is more efficient and whether they are intrinsically worth the potential destruction – as evidenced by the Meta and Amazon incidents – while multiple studies show that forcing workers to use AI tools may actually make their jobs harder.
More on AI: Crooked new AI tool “clones” software so original creator doesn’t have copyright on new version
