I agree completely. But - and this is with my scientist hat on - without quality / productivity measurements it’s very hard to know, or prove, when and how each tool is useful, and best applied. Your point about pair being better with co-located teams is a great example: it makes sense, but can a team actually measure quality with co-located and non co-located pairs, and then see if there are ways / tools to improve effectiveness overall? With a well automated development process it should be possible to track and measure these things, and identify key factors that yield better code, and better products.