Things that can scale
One of the most mistakenly-understood ideas in the startup world is the idea of doing things that don't scale. The original article by Paul Graham focuses on some tricks that early entrepreneurs followed in order to be efficient later on.
Sadly, most people keep only the title in their minds, and loose focus on what is really important: if you take venture capitalists money, your project should scale.
For example, if a company does data analysis, with a global scope, let's say drug discovery, but every project has to be manually developed, tuned, and deployed, it will never truly scale. It may be OK for a consultancy or a custom solution provider, but it is not for someone aiming to have a repeat business case: sell thousands of copies of the same piece of software.
And that is when the real distinction happens: Starting with something that does not scale (like going to single customers and deliver custom solutions) can be good for understanding the business case and for convincing the first few champions. But if there is no common case, a one-size-fits-all type of scenario, then there's no way the company will scale.
Doing things that don't scale also exposes entrepreneurs to artificial dead-ends. Such is the case of Theranos, for example. Using an old technology to validate the business case can be a good idea. This is what Uber did, pairing drivers and passengers by hand. But then, if the technology does not scale, if there is no new technology (like with Theranos) or if there is no way to automate the process, then the business is doomed.
These are the other notes that link to this one.