The main point I got from this article was the necessity to change "how wet do things" to make them compatible with AI.Where I work, information is spread over a lot of different systems: Slack, Confluence, DevOps, SharePoint, a separate support/ticketing system. It's easier and less time consuming for me, following my current process, to write descriptions, acceptance criteria, and release notes from source material, and then just get AI to tweak the wording a bit, than it is to compile it all into a single "unit" for AI to then do each stage individually as a first pass.
If our documentation was more centralised / organised on a single system, then it would work much better. But that migration ain't never gonna happen in a reasonable time frame.
I'm working on an ideas or two to compile things temporarily, but that also takes additional time.
All these things take time and planning, which is rarely factored in. Just do it! Just use AI!
I should also point out that I know a teacher who uses it to greatly improve productivity in writing reports: Give it the main points that should be made about a child's learning, then let the AI write the words that join those bits together. The results have apparently been great, and has taken a fair bit of stress out of the report writing process.