But XML logic sucks for autopilots... A complex AP is much easier to do via nasal.
To me it least it seems reasonable to let Nasal do guidance and AP/JSBSim elements control.
Guidance tasks (like 'pitch to +10 degrees' or 'hold -30 fps vertical speed') don't need to be computed or updated at a high frequence (every few seconds is usually sufficient) and the can be fairly complex as they might involve optimization processes over fairly tough problems ('assuming I hold current pitch and thrust, where will I be within the next 30 seconds given gravity, current wind and the fact that I descent into more dense air ')
Control tasks (like 'how to move the airfoils to pitch to +10 degrees') need to be done at a high rate to avoid instability - you basically need them to run at FDM rate of 120 Hz or you will get nasty oscillations.
So the division of labor along those lines will give you a lean and stable AP, and anyone who believes that you can code optimization over trajectory prediction scenarios in JSBSim as well as in Nasal has likely never tried either.