I know the NASA thing was meant to merely underscore a point but I've seen this comparison so many times that it bugs me and I wish folks would stop using it. The average NASA sw engineer develops a ridiculously low number of lines of code a year (I think I read something like less than 100 ines). A 6000 line patch has 2500 page spec.
The dynamics are vastly different than in any commerical software.
A nice article on the process is here http://www.fastcompany.com/online/06/writestuff.html . Makes a good read.
I've interned at the NASA software testing facility (IV&V) for a couple of years. Their practices for V&V are really quite amazing and thorough, and even more so for validation of unsupervised neural networks.
It very much puts to shame conventional software practices for software assuredness. The fast company article mentioned seems to only glance at the level of complexity they employ to consistently obtain correct software.
I thought this was a good article but one of your points touches on one of the reasons why purely engineer-driven start-ups fail: no business sense. I agree that the MBA-types need to stay away from the engineers on a day-to-day basis, but I've seen too many start-ups shoot themselves in the foot because they launch a product without any market research and no understanding of how to make partnerships that will help them scale.
...why purely engineer-driven start-ups fail: no business sense.
I'm not sure how you managed to draw that conclusion. His bonus point is that the CTO didn't even do work, that's hardly engineer-driven. And they had a Stanford MBA, who was apparently a drain during product development.
Most successful technology startups are engineer-driven. Google is the ultimate example of just how far a couple engineers can go without any "business sense".
What does it mean to invent a product? Adwords is just a name. The actual system was built and rebuilt and improved many times. Eric did do a lot to make it a success though -- he's part of the reason why Google makes so much more money than Yahoo.
I know the NASA thing was meant to merely underscore a point but I've seen this comparison so many times that it bugs me and I wish folks would stop using it. The average NASA sw engineer develops a ridiculously low number of lines of code a year (I think I read something like less than 100 ines). A 6000 line patch has 2500 page spec. The dynamics are vastly different than in any commerical software. A nice article on the process is here http://www.fastcompany.com/online/06/writestuff.html . Makes a good read.