The way I've been putting it to people lately is: Never underestimate how hard a problem can grow by making it big. And also, at times, it is hard to appreciate how difficult something becomes if you haven't walked the path at least partially.

Like, from work, hosting postgres. At this point, I very much understand why a consultant once said - "You cannot make mistakes in a postgres 10GB or 100GB and a dozen transactions per second in size". And he's right, give it some hardware, don't touch knobs except for 1 or 2 and that's it. The average application accessing our postgres clusters is just too small to cause problems.

And then we have 2 postgres clusters with a dataset size of 1TB or 2TB peaking at like 300 - 400 transactions per second. That's not necessarily big or busy for what postgres can do, but it becomes noticeable that you have to do some things right at this point and some patterns just stop working hard.

And then there are people dealing with postgres instances 100 - 1000x bigger than this. And that's becoming tangibly awesome and frightening by now, using awesome in a more oldschool way there.