METR proposes 'expenditure horizon' metric for AI agent cost

METR introduced a new metric called the 'expenditure horizon' that puts a dollar figure on when AI agents become more expensive than human workers for a given task. Early results on the NanoGPT speedrun benchmark show agents underperforming on cost-effectiveness, though METR notes the metric has known blind spots.

Why it matters: This gives a concrete economic yardstick to a debate that's mostly been argued qualitatively: whether agentic AI is actually cheaper than hiring people once compute and failure costs are counted. If newer model generations shift the picture, as METR suggests, this metric could become a standard way to track the real ROI crossover point for deploying agents versus humans.