McKinsey & Company warns companies that AI spending may rise even further as agents become more common.
Compared to text-based AI tools, agents can cost much more to run because they are actually completing tasks. Such tasks are often multistep processes, so the same goal can be accomplished in a variety of ways, and costs vary widely. According to a McKinsey study, the cost of completion for different agents could differ by as much as 30 times.
“Imagine like you’re running an operation,” said Lari Hämäläinen, a McKinsey senior partner, “but every day there’s a 30x difference in the cost.”
The consulting giant found that about a third of organizations spend more than 10% of their technology and communications budgets on AI in its 2026 State of AI survey. McKinsey said 60% of survey respondents planned to increase their AI spending next year. And about one in five said that AI spending was beginning to create constraints in their operating costs.
“So the spend is now becoming quite material and visible,” said Tanguy Catlin, a McKinsey senior partner and director of the McKinsey Global Institute, on Tuesday.
The issue is particularly acute for software-development teams using agents to automate coding, “which obviously is very token hungry,” McKinsey said.
The warning comes after companies spent much of the past year encouraging workers to use more AI. Amazon shut down an employee-created leaderboard tracking AI token use after some workers performed tasks solely to climb the rankings. Companies, including Coinbase and Salesforce, have also begun putting limits on AI use as bills rise.
In a report published earlier this month, McKinsey said agentic AI can cut the amount of time humans spend on certain transformation-office tasks by 35% to 40%, and sometimes by 70% or more. It argued, however, that companies need to measure whether their agents are producing enough value to justify the cost.
“A lot of this we are only now learning,” Hämäläinen said. “In the next 12 months, it will become very relevant.”
