The machine performs half of 12 boring jobs at the bank needed every month. But there is still another half dozen jobs the IBM still cannot do. It doesn't have the punch cards. So the roomful of smokers does the rest of the jobs in a couple of weeks, ready just in time to start over with next months tasks.
Should IBM make the machine or its programs twice as fast, or have twice as many features, even if the programs are half as fast? They should double the amount of features. And double it again, and again, and again...
But scale those "close to nothing" downsides times 7billion people, 100k seconds in a day, 365 days a week. And suddenly, the ammount of time and electricity alone wasted on something like, waitiing for windows explorer to start, adds to real world scales.
...But also, this implies that optimizing is inherently costly. Which is not always the case. Just the culutre of 'not caring' makes things go slower then they could, while still using the same high level languages etc.
But then you talk to end users and it turns out they work 80 hours on one big document with input lag of several seconds. Some actions can take 15 seconds. It’s mad. It’s larger than anything we imagined anyone to work on. It must be infuriating. And they don’t get beefier machines despite it being an instant cost saving. Because software is seen as magical anyway. They do in two days what would take 50 days to calculate by hand! That’s fast! So when you ask them what to prioritize surely they say “make it faster with heavy documents”? No, they want more features. Because a feature can save then days or weeks or months. A faster program just saves them hours. So I sort of understand them (and why we build software that grows slow).