True but also there has never been a better time for total rewrites of legacy code. One could even hope it's the time COBOL finally goes out of fashion in some industries.
On the other hand, models a few years from now will be smarter and more likely to have a track record of success with this task, so being one of the first to do it makes no sense even it had advantages.
So there's no urgency to rewrite; it's all risk with little to no reward for a problem that will probably be much easier to solve if you defer it a little longer. Like, first at least get to 100% AI automatic development for all new products/services, and then for all maintenance, so you know that it actually works.