I think that's the distinction James Kwak is trying to make. He isn't saying that the business processes (in this case, buying and selling stocks) was especially clean before software. He is saying that software makes the system faster (amplifying the effects of a poor business process, as you say) while removing a human buffer. When a human order-taker sees an unexpected zero on an order sheet he or she has enough common sense to ask the person handing it to him or her whether the figure is correct. Most financial software doesn't have that "common sense", so it's easy for inadvertent slip-ups or programming errors to cascade into firm-ending chain reactions.
Obviously, the solution to this isn't to go back to slow and inefficient human processes. I think the solution is to continue doing what we're starting to see today - software that asks whether we're sure we know what we're doing when it sees a variable that's out of line. In addition, I think that many software systems are way too monolithic. That is, every component of the system implicitly trusts every other component. I would like to see software broken up into smaller communicating components, where each component can make assertions about the input and output of other components, and components can fail without bringing down the whole system. The system I've seen come closest to this is Erlang's OTP, but there's no particular reason for that model to be limited to Erlang.
Tell this to Mizuho Securities, where a human sold 620,000 shares at 1 yen instead of 1 share at 620,000 yen. ~$300 million loss in a single botched trade. Humans at both Mizuho and the Tokyo Stock Exchange saw the trade and all declined to intervene because they assumed they had insufficient authority to countermand the trader.
There's a very good reason why the enterprise software industry sells billions. Because they save 10s, 100s or 1000s of billions.
For custom software, that's what a good consultant/PM is actually for, to translate 'what we thought we were doing' into 'what we're actually doing' into the final product of 'what it should actually do'.
I once worked for a company that sold project and document management software. The strangest call I ever got was from a small 20 person company phoning us up and saying 'What is Sue supposed to do now, she doesn't have anything to do'. That's what you bought the software for, to save money. Added bonus, management gets massive strategic oversight. But in the end it's up to them to use it.
Not just one company. IM apps alone may have saved trillions.
Compare to the water cooler, where it might take me a whole minute just to walk down there. Then we sit around and shoot the shit for maybe five minutes. Total time spent, let's call it seven minutes during which nothing got done except some simian grooming.