Having seen some spreadsheets in my time, I can imagine what the training data is like.

My agents can deal with Excel all right. What’s the story?

The story is that the writer is Orcaset, which is trying to sell you their own analysis tools.

Yup. Their benchmark appears to consist of giving an agent XLSX files that use advanced Excel features, but no copy of Excel, and expecting the agents to interpret the spreadsheet the same way Excel would. So the agents try LibreOffice instead, and if that doesn't work, they try to write Python scripts. But the behavior of the Python doesn't match the behavior of Excel.

To put it politely, this seems like a self-inflicted problem. If you really want your agents to interpret advanced Excel features exactly the way Excel would, have you considered maybe giving your agents Excel?

I've tried to 'reverse engineer' some of our excel sheets, and it's difficult to get the model to stop scolding me for using so many =INDIRECT statements.

I once tried using Gemini directly in Google Sheets to make a graph. I never managed to get it to do exactly what I wanted; after wasting way too much time on it I was able to fix the problem in a matter of seconds. It's fascinating because it is quite different from the situation in coding agents where they are in general really quick to read code and find issues/fixes.

The key is to have agents generate code which exports Excel with live formulae (there's a good Python library).

There are way worse choices to fall back on than LibreOffice. Heck, it's the primary tool I actually use in my business. I can tell you first hand that it does not handle all the random stuff that corporate Excel users come up with, but if you start with basic principles the files usually work great in the other direction.

Building financial models in python will never work

The end user is not a software developer

False.

Have your agent call DuckDB to load it first.

I didn't know that. Agents can't do today what I used VB for 20 years ago?