Unlocking agentic value: a new investment discipline for the agentic era
EY's analysis establishes that agentic AI requires a fundamentally new investment discipline, distinct from traditional software or generative AI budgeting, because token consumption in multi-step autonomous workflows scales non-linearly and unpredictably with task complexity. Enterprises deploying agentic systems face a new cost unit—the token—that accumulates across reasoning loops, tool calls, and context windows, making per-task economics difficult to forecast under legacy IT spend models. EY argues that CFOs and CIOs must build token cost visibility into business cases from inception, treating inference spend as a variable operational cost rather than a fixed capital outlay. Without active token governance—including model selection, context pruning, and workflow design discipline—agentic deployments risk eroding ROI even as they deliver automation value. The piece frames token cost management as a board-relevant strategic capability, not merely a technical optimization, as AI inference spend scales enterprise-wide.