In the case of Microsoft Office native desktop applications the LLMs can figure out everything underneath through a mix of decompiling the binaries, reading the product documentation, and exploratory testing. Some SaaS is more opaque.
Functionality piles on top of each other, products are such a garbled mess with fixes built on top of shaky grounds all over.
20% of software is code, 30% is design/strategy/marketing/telemetry/usage, rest of the 50% is lots and lots of stuff baked into software from user feedback over the decades.
Any software product that isn't in a similar balance is just crap to use, for instance Windows it's 20-30% code, 60-70% marketing/ads and the rest 10-20% bug fixes..
There is 0 user feedback being addressed.
In my opinion, AI is the thing that should be a wake up call for software companies to actually build on top of feedback proactively and with sharp intent.
Good luck and please notify us about the progress.