> 'But only I can fix that complex race condition, that popped up half a year after development' - well if it was good enough with the bug for people not to notice it for half a year, it's going to be good enough for another half, until the new devs can fix it.

From my time at AWS, and in light of the recent DynamoDB dns race condition bug, this is a line of thinking that is problematic.

There is so much complexity in the inner workings of any one of those services. Many of those services are now composed of many teams, each having turned over completely multiples of times.

Yes there are runbooks and knowledge passed down from each generation, to some extent. But when you hit some critical outage, edge case, etc., those people running the systems have such a different relationship to the system having simply maintained it vs. built it.