McKinsey Report Warns Autonomous AI Agents May Drastically Increase Enterprise Costs
A recent McKinsey report highlights that deploying multi-step AI agents can lead to unexpected enterprise cost spikes, with execution costs varying by up to 30 times for similar tasks. Additionally, about 20% of surveyed organizations report that AI expenditures are already exerting pressure on their operational budgets. While AI agents can reduce manual task execution time by 35% to 70%, their autonomous multi-step execution loops consume massive amounts of tokens. This financial reality forces enterprises to re-evaluate AI return on investment and implement usage restrictions, shifting tech leadership priorities toward strict cost governance. Software development teams using AI agents for automated coding face particularly heavy financial pressures due to high token consumption. In response to rising bills and unmonitored usage, major companies like Amazon, Coinbase, and Salesforce have begun placing restrictions on employee AI usage.
## BACKGROUND
Unlike basic LLM chatbots that only generate text responses, autonomous AI agents can plan actions, call external software tools, and execute multi-step workflows autonomously. However, because each step in an agentic loop requires repeated LLM API calls, unoptimized workflows can cause token consumption and overall inference costs to skyrocket.