In case that wasn't rhetorical, you can upload an xslt or csv file to ChatGPT and ask it to do stuff and it'll give you answers and can update the file with new formulas that it writes and you can download an updated copy as an xslt or csv. It's pretty neat, for those who need help writing formulas.
"They will be figureheads with some responsibility, but not day to day operational duties" sounds like the job description of a politician in a democracy. They get hired (sorry, elected) by a mob who knows stuff all about the day to day running of the place and so vote on other issues. As a consequence the people they elect are rarely good at anything other than getting elected. Nonetheless they stand up and proudly claim they will fix everything now they are are in charge, and consequently get blamed for whatever goes wrong during their tenure, and then we fire 'em.
The main danger is when nothing goes wrong. That lets them stay for a while, and if they are actually competent they may start running things, which lets them re-arrange the laws, the police and the judiciary so they are hard to boot out.
If that happens we tend to introduce term limits. We can't have competent managers running the place - that's far too dangerous.
/s
Once AIs are close to doing the job, they have fundamental advantages. They're faster than humans. They have far more network bandwidth. The next frontier is figuring out how to interconnect AIs into a management team.
Can AIs use spreadsheets yet?
I'm still expecting "Microsoft Middle Manager 2.0" at some point.