Disorder and delay are diagnostic terms.
Disorganised was probably used more descriptively to relate to difficulties planning, pretty much as it sounds.
The simplest model of language planning is the Garrett model.
This assumes that we have a message in mind before we begin to speak - an idea of something before we add words to it, e.g. a mental representation e.g. your ds wants you to make him a sausage and he has a mental image of you doing this in the kitchen and it involving fridges and plates and you.
To convey this message to someone else, there is a certain amount of linguistic assemblage that needs to happen. Each individual word needs to be retrieved, then it needs to be assembled into some order, then refined so that it is grammatically accurate and with an intelligible sound "shape" (much more detail about this sort of process here: www.smithsrisca.demon.co.uk/PSYgarrett1990.html).
In your ds's case, this isn't really what's happening. From the limited data I have on his production, I would say that he's not actually retrieving or assembling individual words, but large chunks of language that he has associated with context. This is a "gestalt" style of language learning: it suggests that in terms of language analysis/planning that this process is not adequately functioning.
Basically, instead of working from the message level out to the minor grammatical details needed to execute what he has in mind (bottom up), he works from the end point of having previously associated a long chunk of language with that context. All very well if his message happens to correspond with a chunk of language that "works" and makes sense to the other person, but what he's actually doing is taking bits of sentences that have worked and kind of throwing them together, meaning they only make sense if you, as a communication partner, have specific individual knowledge of the context or work to fill in the gaps (which an adult will do, a peer may not). It is a haphazard approach - a bit disorganised, I guess.
My only issue with this interpretation is that I think, in some ways, this is the norm for a lot of verbal individuals with ASD and at some point, for many, they break the code/pattern, suggesting that they may be developing language in an altogether different way to this model (which is based on data of the errors made by adults with aphasia, who would potentially have a very different linguistic model and have a static, damaged language system vs a dynamic, developing one). However, that's just my gut instinct on the issue.. I haven't yet come across a model that resolves this issue.