I worked on FB integration where a feature being used was disabled and the API replaced with a new one between our product freeze and release (actually the old API was closed a week and a half before the replacement was in place). If I was working at a smaller company at the time, our team would have had no way of knowing WTF happened, other than it didn't work.
The fact is, even with a large corp to corp relationship, you still can't absolutely rely on API consistency... it's pretty sad, but true... Netflix, Twitter, FB, Google and others have changed/revoked APIs before with minor to significant impace.
This is part of the reason why you shouldn't try to make applications designed to scale/adapt for years... design the simplest thing that does the job, and in as modular a way as possible. Avoid "enterprise" patterns as much as possible in favor of the simplest thing that works and is cheap/easy to replace if/when needed. It doesn't mean don't test, consider performance or write good code, just don't rely on it being around, and over-invest in that it likely won't.
You never know... a piece of software I once wrote to test our own product turned into its' own product and pieces of it are still in fortune 100 deployments... It's horrible, and was a hack to begin with. It didn't even support half the specification it was designed for (only the parts we needed)... just the same, it got the job done.