Direct visible effects are wrong decisions entangled in spaghetti-like complexity.
It's hard to reach technical bottleneck in well designed systems. Computers are really fast novadays. They will vary greatly depending on what kind of system it is. Out of resources – cpu, memory, network, disk io – likely the weakest of them will be saturated first – network. But that's not a rule, it's easy to have system which will saturate ie. CPU before.
Competence is expensive :) While I mostly agree, even well designed systems have (sometimes considerable) tradeoffs.
> It's hard to reach technical bottleneck in well designed systems. Computers are really fast novadays.
I have been listening to how fast moderen computers are for the better part of the past two decades, yet as I user I still have to deal daily with too many of slow software and slow web services.
Somebody once said "cheap things are expensive". This idea applies to developers as well. Cheap developers will drive company through bumpy roads towards uninteresting plains. Good developers not only pay for themselves but bring orders of magnitude more cash in. Only thing that touches on this that I can find is "software craftsmanship manifesto".
A lot of people don't see the effects of their decisions. They leave a company after 3-5 years and go and work somewhere else where they get to make the same mistake again. The bottleneck indeed is lack of competence.
As for technical bottlenecks, it's quite easy to hit a wall. Be it through layers of stupid or unexpected success. We have unexpectedly reached the limit of what is possible with x86-64 on a couple of occasions due to stupid decisions made over 10 years previously for which there is now no longer the budget or attention to fix.
Money. The most common bottleneck is money and customers. Use whatever helps you get new customers faster.
Don't be scared of having to make changes in the future. Do the small amount of work it takes today to make sure your transition in the future is easy.
Transitioning from a SQL queue to redis it's only difficult if you have a bunch of SQL throughout your code. If you have that, you did it wrong.
I agree that there are many cases with low workload where that would be plenty.