Hmm
First, the reflow is coming from a company that has progressively reduced the effectiveness of the Acrobat UI over the last 20 years to the point where I can barely find stuff in the latest release that was once obvious. If they're using AI to put things on the screen in the right place, perhaps they could start with their toolbars.
Second, as many have pointed out, PDF is a page description language. It excels at positioning things on the page with exactness, and doing so in the correct color. It was never designed for reflow, and the internal format doesn't really allow this. Even now, any product extracting text from PDF has to guess where lines, columns, or even words end - it's very common for a whitespace to be represented with a simple "move cursor" operation rather than any sort of semantic space. PDF text predates Unicode, and it's common to find documents without the requisite structures to identify the meaning of the text. I expect this effort is doomed to reside somewhere between barely-acceptable and outright failure.
Someone pointed out earlier that the most obvious reflowable document format - HTML - would be a likely contender. You're not the first to raise this, is all I can say. Maybe one day Adobe's wheel-invention subdivision will see the light. Mercifully PDF XMP is no longer with us as of PDF 2.0, so the path is clear to combine a proper reflowable format (HTML) with a proper page description format (PDF).
Yes, in terms of accessibility it isn't as good as HTML, but that is being worked on. Tagged PDF is very much a thing, and if a document is to be reflowed I would personally be starting by making sure it has the necessary information to do this, rather than relying on the wonders of AI. That means tagged PDF, aka accessible PDFs, aka PDF/UA. If you're generating PDF, do yourself, your customers, search engines and posterity in general a favour and switch to PDF/UA.
Very, very anon.