Handling simultaneous edits, guaranteeing correctness, handling massive data sets… as amazing as AI is, this is a huge can of worms.
Any nontrivial excel sheet should be a database backed application, the finance guys who build them are in every sense programmers they're just using the wrong tool by convention.
* Redo UI patterns (for old users): Only matter for on-boarding and compatibility, and probably need to redo the anti-patterns, not necessarily a good idea, or at least must be hold before the next:
* Define a much better internal model than excel has (I bet a mixture of relational, array model as core, then a reactive layer on top), including proper support for correct datatypes (not localized!), dates in all shapes, currency, commodities, unit of measure, etc, tables default as fixed type per column instead of free-form, then we have proper "variant" values, use algebraic types, and so on...
* Then redo the UI
* Define how import/export correctly (so not more non-sense about import based on localized dates, currencies...)
* ... and from legacy formats like Excel
P.D: I explore the idea on https://tablam.org and my goal is eventually marry what excel+access do in one app
So long as you don't have things which aren't dates but which Excel insists totally are. (Meme: https://knowyourmeme.com/photos/2429759-involuntary-celibacy...)
But yes, cans of worms are many, and Tardis-like they are larger inside than out.
Let the dinosaurs go extinct!