As in : it doesn't work, no one knows why, no one can fix it and no one has a plan that doesn't involve starting from the beginning and doing it differently.
consider two programs
Good prolog :
on(a,b).
on(b,c).
above(X,Y) :- on(X,Y).
above(X,Y) :- on(X,Z), above(Z,Y).
->above(a,c).
You'll get TRUE (yay) a is above c ! GOOOOOOOD prolog.
but now :
on(a,b). on(b,c).
above(X,Y) :- above(X,Z), on(Z,Y).
above(X,Y) :- on(X,Y).
-> above(a,c).
ERROR: Out of local stack
Exception: (1,675,220) above(a, _4808) ?
ERROR: Out of local stack Exception: (1,675,220) above(a, _4808) ?
Prolog gets stuck expanding the first search until it can't carry on. One line is out of order in the program and you are dead, and the problem is that you will not find that line until that question is asked, and when you find it (in a real example) it really is the devils own job to work out where the expansion is coming and how to fix it.
But there you go - search based logic resolution, brought to you by 1970's computer architecture (small main store) and solvers. Note, not only is "it doesn't bloody work" an issue, but also (probably much worse) "it keeps hanging for five minutes". This inconsistency and problem is hard to replicate and very, very hard to debug. It's caused by an occasional big search happening in the knowledge base, but a big search that doesn't end up exploding the system, it's often accompanied by shifty developer behaviour involving parameters and special cases. In a real project it's a nightmare.
Now : NOOOOOOO! Instead use Problog or something similar and reason using Answer Sets. For your first steps an answer set system looks very like Prolog, but does not suffer from the issues above (as far as I can see). There has not been (so far) a big charge to use this in prod (with Prolog and KBS's there was, it failed, because above) because everyone is playing with 40 layer 200 billion parameter neural networks and wondering why proving they can beat a tiny part of one benchmark or mostly steer a car without killing many people is not translating into application success. And this is a good thing because I don't think that the tech is quite there yet and the community needs five years and about half a dozen big "let's make this practical" applied research projects (btw. keen on partners?) to actually make this manageable for people with deadlines and investors.
But - it's going to help you more than Prolog.