I don't agree fully. My understanding of the Google level scale is that you are forced to deal in good abstractions. You need to pick the right interface, nothing can ever be responsible for more than one thing, and you need to make things extremely composable. With a foundation like that, you get the ability to model your system using even simpler substitute components, making it possible to reason about behaviour in conditions which are difficult – if not impossible – to actually test.

If the ideas sound familiar, that's because they're basically just good software design. Except if you don't design things well from the start, you have to either kill it off or spend years fixing it.

I think a lot of this could be the exactly right mindset for many of us. Sure, there's something to be said for an MVP, but with some experience in designing composable abstractions, you can create an MVP that scales up to real workloads in half the time of your competitors, who basically have to rebuild their thing from scratch.