McKinsey warns of ballooning AI costs as agentic workflows scale

The scale of the challenge is already manifesting in corporate balance sheets. According to McKinsey’s 2026 State of AI survey, one-third of organizations now allocate over 10% of their technology and communications budgets to artificial intelligence. With 60% of respondents planning further spending increases next year, one in five firms report that these investments are beginning to create genuine fiscal constraints.

Software development teams are particularly vulnerable, as agents tasked with automating code generation prove to be exceptionally token-hungry. This surge in consumption has prompted some enterprises, including Coinbase and Salesforce, to implement hard usage limits to curb runaway bills. McKinsey senior partner Lari Hämäläinen noted that the lack of consistency in operational costs creates a new layer of management complexity, as a single task can fluctuate wildly in price depending on the execution path chosen by the agent.

While McKinsey reports that agentic AI can reduce time spent on transformation-office tasks by up to 70%, the firm cautions that these efficiency gains must be rigorously measured against the rising cost of compute. As Tanguy Catlin, director of the McKinsey Global Institute, observed, AI expenditure has moved from experimental to a material line item, forcing leadership to justify the value of every automated workflow before the bill becomes unsustainable.