Here's the deal, you need to use both: (1) a bottom-up Agile, and (2) a top-down estimation approach. You need to do them at different times and for different

Here's the deal, you need to use both: (1) a bottom-up Agile, and (2) a top-down estimation approach. You need to do them at different times and for different reasons. The top-down estimation at the start of a project needs to be done to the lowest level of detail necessary to get a Go/No-Go decision made. You can't realistically start an Agile-managed project without understanding where $10k is in the ballpark of $10M. That's just foolish. This top-down estimation needs to be done to the level-of-detail and accuracy required by your Go/No-Go decision. Then, on a day-over-day basis, you use Agile to manage the actual project operating process. (I'm speaking loosely here, of course, as long as Agile makes sense.) This allows you to be dynamic week-over-week and respond to outside forces, new information, new technology, etc., within your budget. What people miss is that no sensible project sponsor is going to accept, "we'll finish it when we finish it for however much it costs" unless it's pure research. And if I only need accuracy to the closest $10M, then we can keep the estimates high-level, if I need it to the closest $10k, then we need lower level estimates. It's just the nature of it. As a project sponsor, I need to know what's possible.