Agreed. But and then... making small stories only works well for a limited horizon. So, now your estimating only works for planning ahead a limited time. You use that to build a general confidence in having a constant stream of feature output. And the same people can then make some assumptions about future projects, based on similar projects in the past. Which works great for feature factories, less so for committing to a tight, hard deadline for a big project that exceeds the limited horizon.