It seems that the tr ftp and the traditional definition of ftp does not always align.
Within the tr ecosystem (indoors only as there is currently an outdoor ftp bug? But that might be predictions only, not positive) the ai training number seems to do a good job of assigning doable and progressive workouts.
You could do a long form test to get the traditional ftp value but it would probably make sense to leave the ai number as is for training purposes unless you have a specific reason to change it.