I think the most important part of this would be an ML model that generates new office jargon on a regular basis. It seems like this should be easy, since it's mostly just remixing existing language in impenetrable and quasi-meaningless ways.