alt Hacker NewsI actually talk about this in the article. Merely using files is how we got out of the dependency on MS Office. Multiple efforts reverse-engineered them, including Google Docs. Yes they were undocumented, but as long as stuff has to be stored on the disk under user’s control, the overall dynamics are very different from you-can’t-see-the-files systems.
Is it really so different in the service context? You could also reverse engineer the HTTP endpoints and formats used by Word Online to export data from the service. It doesn't feel all that different, except perhaps that the online service can try to detect your custom tooling and block you, whereas with static data that isn't possible.
But other than that admittedly real extra problem, the bulk of the work would still be understanding the undocumented protocols and feature semantics, then matching the feature set.