No, I’m saying that JPEG XL can perfectly represent old JPEG/JFIF data, so on the server side you can store all your image data once and more efficiently, and still support old clients without any lossy cascade or the CPU load of having to re-encode. That is what is meant about it offering backwards compatibility.

It’s “compatible” in that it can represent old JPEG/JFIF data more efficiently and in less space, and the transformation to JPEG XL and back to JPEG/JFIF is lossless (in that you don’t lose any /more/ quality, you can get the same bits back out) and quick enough to be doable on-demand. You could, for example, re-encode all your old photos on your CDN as JPEG XL without loss of quality but save a bunch of disc space and bandwidth when serving to modern browsers, and translate dynamically back to the old format for older browers, all with no loss of quality.