For the few startups that do experience significant growth, it's much easier to find talent (engineering and otherwise).
This is why a product I was working on almost got killed. The senior engineer working on it wanted perfection before delivery. The project was already one year overdue. He suddenly came in one day and rushed me to finish up something and he wrapped up a couple things and we launched at the end of the week.
Found out that the project was going to be axed if we didn't deliver very soon.
I agree, but this is mostly true because there's so much low-hanging fruit out there. If you look at the current successful startups like Uber and Airbnb, they're really trivial applications. That doesn't mean they can't create huge amounts of value by "trivially" repackaging services and connecting people.
Once the low-hanging fruit is developed during the next 10 years or so we can finally start focusing on actual high-tech engineering startups.
(Soylent doesn't even make the stuff. That's outsourced to Jasper Products in Joplin, MO.[1])
Very few things require genuinely novel pieces of code to work. Yes, lots of value can be captured using code, but you could totally start something like airbnb with a lot of manual labor clicking in a GUI.
Djikstra said "Beware premature optimization" and that's quite true. I think automation should be used when things hurt to bad to do manually anymore. For some people, that's any task which must be performed twice and then you write a script. For others it might be years manually updating excel for bookkeeping every night. One can go real far manually updating excel...