India, Aug. 21 -- AI inference costs for agentic workflows are expected to increase more than fivefold through 2028, even as the cost of individual AI tokens continues to fall, according to Gartner. The research firm says the growing complexity of AI applications is creating an "inference paradox" for product leaders, with improved unit economics potentially driving higher overall AI spending.

The shift from AI assistants and chatbots to systems capable of executing multistep tasks is changing the economics of AI deployment. While foundational model costs are declining, more sophisticated applications require models that can reason through tasks, use tools, evaluate outputs and make decisions across multiple steps.

Gartner said these wo...