Plenty of documentation, and plenty of code that the AI can read itself.
E.g. if a library has a bug that has a common workaround, it can learn that from open source code using the library that uses the workaround.
That assumes there is documentation or examples. A big reason Stack Overflow took off was people struggling with things like the Android API documentation.
Some of those discussions made people go figure out how to do it, and then post it as an answer. The knowledge didn't exist anywhere until they did.
The only way I could see this being surfaced the same is if the code essentially had a SO answer written into the doc comment.
lots of undocumented gotchas that only surfaced because someone used it and posted about it
This and the the other thread that talks about RL and synthetic data seem to suggest that AI can figure out all the technical issues without humans looking into them. I'm not sure if that's true at all.