McKinsey's guide to agentic-workflow economics shifts the unit of analysis from the number of agents to the economics of a concrete job.
That framing forces teams to count the whole path: labor displaced or augmented, model and tool cost, exception handling, review, delay, and the value of a completed outcome. A fast agent step can still make the workflow more expensive if it creates a larger verification queue or fails on the cases that matter most.
A useful pilot should therefore start with one bounded workflow and a baseline. Measure completion rate, cycle time, human intervention, rework, risk, and unit cost before treating deployment volume as evidence of value.