Cost versus value: managing the performance of agentic AI systems

By McKinsey & Company (Dan Tinkoff and Lieven Van der Veken, global leaders of QuantumBlack, AI by McKinsey).

This McKinsey Quarterly mailing features a piece titled "Cost versus value: Managing agentic AI system performance", which looks at how companies should measure what they actually pay for generative AI, and especially for agentic AI systems.

The main idea previewed in the email is that price per token no longer works as a metric for the real cost companies take on when they deploy generative AI. According to the teaser, the piece explores this in a conversation with David Tepper, CEO of Pay-i, who goes into detail on the new measurement economics said to be emerging for assessing the spending on these systems and the value they deliver.

The email doesn't go into further substance in the body of the message. It is a promotional email linking to the full article on McKinsey's website, so no further specifics (figures, examples or particular findings) are available without opening the original article.

The newsletter also has an "Also Consider" section with two related pieces: "Frontiers of compute: The technologies to reduce AI inference costs" and "Where AI will create value—and where it won't". They cover related topics: compute infrastructure for inference and where AI will create value.