McKinsey Warns AI Agents May Cost 30 Times More Than Chat AI

Summary

McKinsey warns that the spread of AI agents could push enterprise AI spending substantially higher because agents execute tasks through multiple steps rather than only generating text. Its research found that different agents performing the same task can have costs that vary by as much as 30 times, making agent selection and task design important drivers of the final bill. The firm’s 2026 State of AI survey found that about one-third of organizations already allocate more than 10% of their technology and communications budgets to AI, 60% plan to increase spending the following year, and nearly one-fifth say AI spending is beginning to weigh on operating costs. Software development teams using coding agents are especially exposed because automated programming can consume large amounts of tokens. Some companies have begun restricting usage after incentives such as internal token rankings encouraged employees to submit unnecessary tasks. McKinsey also reports that agentic AI can reduce labor time for parts of transformation-office work by 35% to 40%, and sometimes by 70% or more. The central business question is whether those labor savings will exceed the agents’ operating costs, an issue the firm says will become increasingly important over the next 12 months.