As companies have just got AI under control, consulting giant McKinsey has handed them a bill that might raise eyebrows. In its latest warning, McKinsey cautions that as agents become more prevalent, corporate AI spending is likely to keep rising — and this time, the cost-cutting logic is completely different from that of previous chat tools.

Compared to text-based AI tools that only generate text, agents need to actually complete tasks, which can significantly increase operational costs. These tasks often involve multiple steps, and there are various ways to achieve the same goal, leading to huge differences in costs. According to a McKinsey study, the cost difference between different agents completing the same task can be as high as 30 times. In other words, the choice of agent and how it is used directly determines whether the bill is moderate or explosive.

This pressure is backed by data. A McKinsey "2026 AI Landscape" survey found that about one-third of organizations have already spent over 10% of their IT and communications budget on AI; 60% of respondents plan to continue increasing spending next year; and nearly one-fifth admit that AI expenses are already starting to weigh heavily on operating costs. McKinsey Global Institute Director and Senior Partner Djordje Katic points out the trend: "So, the scale of AI spending is becoming quite substantial and increasingly obvious."

The software development teams that use agents to automatically write code are the most vulnerable to these bills — automatic programming is naturally a token-consuming machine. Irony lies in the fact that just before McKinsey issued its warning, companies had been encouraging employees to use AI for most of the past year. Amazon once created an internal token usage leaderboard, but some employees were solely focused on climbing the rankings, so the company eventually removed it; when the bill started to rise, companies like Coinbase and Salesforce began imposing restrictions on AI usage.