Source: Fortune
Summary
McKinsey partners Tanguy Catlin and Lari Hämäläinen discussed the paradox of cheaper AI models and rising enterprise costs during a virtual session. They noted that while AI capabilities have become significantly more affordable, businesses are consuming more AI-driven work, especially through autonomous agents. Hämäläinen highlighted that AI vendors are capturing some of these efficiency gains through higher margins. The session also outlined strategies for managing AI costs, including better visibility, workflow optimization, and sourcing discipline. Catlin emphasized that cost management should focus on high-return AI investments.
Our Reading
The numbers tell one story.
AI models are cheaper, but usage is up.
Agents are doing more work, driving costs higher.
Vendors are keeping more of the savings.
Companies are still learning how to manage the economics.
Author: Evan Null
McKinsey on AI Economics
McKinsey senior partners Tanguy Catlin and Lari Hämäläinen addressed the growing disparity between AI cost reductions and enterprise spending during a virtual session. The discussion centered on the firm’s State of AI in 2026 survey, which highlighted a key trend: while AI capabilities are becoming more affordable, businesses are using more of them, especially through autonomous agents. Hämäläinen noted that models with GPT-4-level performance can now be served at a fraction of the cost, but the volume of work being handled by AI has increased dramatically.
Cost and Consumption Paradox
The paradox of cheaper AI models and rising enterprise bills was a central theme. Hämäläinen explained that while the cost per unit of intelligence has dropped significantly, the amount of work being done by AI has exploded. This trend is particularly evident in areas like software development, where AI agents can generate far more code than human developers. He also pointed out that AI vendors are benefiting from these efficiency gains, capturing some of the cost savings through higher margins.
Managing AI Costs
McKinsey outlined three key areas for managing AI spending. First, companies need better visibility into which use cases and agents are driving costs. Second, they should optimize workflows by matching model complexity to the task, caching context, and limiting unnecessary tool calls. Third, sourcing discipline is critical, including managing licenses, negotiating terms, and avoiding over-reliance on a single model or vendor. Catlin warned against indiscriminate cost-cutting, emphasizing that companies should focus on high-impact AI investments.
Agent Economics
One of the biggest challenges in AI economics is the variability of costs when using agents. Hämäläinen noted that the same task can cost up to 30 times more from one run to another, depending on the path the agent takes. This variability is driven by the reasoning and refinement processes behind the output. He suggested that companies should evaluate agents at the task level, considering factors like cost, success rate, and the time required for human verification.
Future of AI Workflows
The session also touched on the need to redesign workflows to take full advantage of AI’s capabilities. Hämäläinen said that the bigger challenge is not just reducing costs but rethinking how work is structured to free up human capacity. He provided a rule of thumb: an agent makes sense if the time to verify its output is a small fraction of the time it would take a human to complete the task. This approach helps companies determine when AI can start delivering value.
