We’ve already seen 90% reduction in intelligence costs multiple times over the last few years. Each time, token consumption rises 20-100x