I may have to recheck on the Constellar Xyz. If I see it happen again, I'll note it. If not, just chalk it up to me going from flawed memory.
It looks like First of the Dragons is even more of a problem card for the AI than I thought, since it also seems to mess with the attack logic. It doesn't even behave as if the AI simply doesn't see it as not being destroyed by battle. Frequently, the AI will launch an attack against First of the Dragons, clearly intending to use a stat-changing effect during the battle, then not use that stat-changing effect, leading to pointless suicides. What appears to be happening (from the standpoint of someone who is just observing the behavior, at least) is that when it launches the attack, the AI doesn't realize that First of the Dragons won't be destroyed by the battle, but at the point where the AI would activate the stat-change effect, it rechecks, sees that First of the Dragons can't be destroyed by the battle, and therefore doesn't use the stat-change effect.
Honestly, First of the Dragons seems like it just may have to be a "special case" for the AI. Basically, make it a higher-priority target for the appropriate S/T effects, blacklist it for the appropriate monster effects (preferably with an exception for if First of the Dragons' effect is currently negated), and code for it in the attack logic (probably the hardest part).
Incidentally, the deck I've been testing with that includes First of the Dragons also includes Evil Hero Dark Gaia, and the AI seems to handle that just fine just as a result of its general logic. It might not be a bad idea for the AI to give Evil Hero Dark Gaia a slightly higher priority for effect negation (since negating Dark Gaia's effect makes its ATK 0), but considering that Dark Gaia isn't used all that often (and when it is, it's usually as a finisher anyway), that may or may not be worth the effort. Come to think of it, this might be something that can be done semi-generically, depending on what information the AI has access to -- if the AI is able to see that a monster's printed ATK is ?, it should be possible to code things so that it will "know" that negating that monster's effect will make its ATK 0.
On a much more positive note, I will say that the ABC deck seems to have come out very well. I haven't yet seen it have the opportunity to pull off the Karakuri play on me, but I can say that, from what I've seen so far, it does a good job handling ABC itself, ABC's pieces, and the Transmodify play.