Large scale reliable systems are antinomic with “launching a startup”. You’re going to go through 2 or 3 pivots and as many refactors, large scale is the last thing you want to optimize for.