Tom Cunningham surveys metrics for comparing AI agent capability as performance changes with expenditure and relative to human performance.