Regardless of what you think about the 37 Signals folks one bit of their advice has always struck me as being particularly insightful and powerful:
Sell your byproducts.
There are two very strong reasons to do this. First, it can be an excellent business. You've created something to solve some problem or remove some pain that your company is feeling, it's extremely unlikely that your company is so unique that it's pain isn't shared by other companies. And it's unlikely your internal tools would be of no interest or utility to other companies.
Second, internal tools are typically of terrible quality. There are various reasons for this but it's a very common pattern due to fundamental pressures and incentives. By selling internal tools you force them to have owners and you force them to have a quality sufficient to be acceptable to the market. This generally vastly increases their quality, which provides a benefit to everyone who uses them, including you.
Except it has been reported by Amazon insiders that AWS is not a byproduct of Amazon's other work but a very much stand alone product with it's own dev group with little connection to devs working on Amazon proper and that Amazon was relatively slow in adopting AWS product for implementing Amazon's internal software.
Amazon always tried to give the impression that they use AWS but (at least in early years) those were half-truths (i.e. statements vague enough that if you squint at it one way you can read "Amazon's internal systems are build in AWS services" and if you squint at it differently, they might just as well say "AWS's servers are hosted in the same data center as Amazon's servers"; public perception was the former and the reality was closer to the latter).
That, of course, doesn't mean that 37 Signals' advice isn't good, just that this is not an example that confirms that advice.
AWS was built in the same datacenters by the same IT engineering team that provided core hardware and networking to the retail website. The top level software APIs were (obviously?) not created before the Amazon retail website.
Amazon IT and AWS invented a virtual machine infrastructure that Retail used as soon as it was available. AWS services were quickly adopted by many teams as well.
AWS has always beeywn the future of Amazon Datacenter IT. Retail is a huge legacy system, so of course Reddit could launch on AWS faster than Retail could fully migrate.
AWS is an externalization of core Amazon technology. Don't get hung up on the fact that Amazon runs on a babysat patched alpha version released before the 1.0-quality version went public. Retail needs some customizations hat your startup doesn't. Really.
This "skunkworks AWS" meme is a story about the business, not the technology.
Yes, this is true. When I first heard about AWS I assumed Amazon was selling excess capacity (because why would an online retailer set out to build something like this?), but in fact, as you say, it was a completely separate project. I've heard from people who used to work at Amazon that they don't use AWS for much there.
I tend to think we use it quite a lot internally. For instance every amazon.com web server has been running on AWS for quite a while now. You'll find more info here.
It should have been obvious for everyone right away that either Amazon was not selling excess capacity or depending on AWS would have been a very bad idea.
If Amazon was selling excess capacity, care to guess what would have happened to peoples services come the christmas rush, or any other crunch time for Amazon?
AWS have always been expensive for capacity you need most of the time compared to renting dedicated capacity elsewhere, as you're not only paying for the capacity, but for enough excess capacity for Amazon not to run into the PR disaster of not being able to handle requests for new instances.
The numbers and plain logic just never matched the perception of Amazon selling excess capacity.
The most amazing thing about AWS was how Amazon managed to get people to see it as the most amazing thing to ever happen for hosting and create the perception that you somehow paid for less, when for most typical web hosting scenarios you pay more for a service that is pretty much promised to be less reliable.
(Note that I use EC2 for some things - it has it's uses, and is a good product, but it was way overhyped)
They didn't have anything internal that had suitable features or quality to sell to the public so they built AWS using the knowledge they'd acquired from building those internal tools. I don't think that's out of the spirit of "sell your byproducts".
Also, in response to pg's post below, Amazon has increased its usage of AWS substantially in the last few years. The main website primarily runs on AWS, many individual services have been migrating to AWS, and S3 is heavily used internally for static storage. It wouldn't surprise me if Amazon was running 100% on AWS within the next few years. Or, you know, so I hear from rumors... <_<
Sell your byproducts.
There are two very strong reasons to do this. First, it can be an excellent business. You've created something to solve some problem or remove some pain that your company is feeling, it's extremely unlikely that your company is so unique that it's pain isn't shared by other companies. And it's unlikely your internal tools would be of no interest or utility to other companies.
Second, internal tools are typically of terrible quality. There are various reasons for this but it's a very common pattern due to fundamental pressures and incentives. By selling internal tools you force them to have owners and you force them to have a quality sufficient to be acceptable to the market. This generally vastly increases their quality, which provides a benefit to everyone who uses them, including you.