How engines have changed opening theory

computers play endgames weakly because pure calculation fails in many endgame positions. this is why databases are used. computers have a major problem with fortresses. they have a horizon effect since many games it takes 30+ moves to convert a win. remember computers calculate EVERY move. Ra1 ra2 ra3 ra4 etc.... which in endgames the positions have too many useless options

This simply isn't true. Computers couldn't calculate nearly as deeply as they do if they looked at every line, unless that wasn't what you meant and I misunderstand.

hence why endgames suck for computers. they try to prune them but endgames are hard to calculate because of certain maneuvering Ra2 might be the best move over Ra1 in a given position but there is no concrete reason why until move 20. In middlegames there are some tricks they use based on looking at forcing moves first etc 

Engines don't always find the best move. I had an engine tell me once when I was sacking an outside passed pawn to divert the enemy king that I should have promoted to a bishop instead of a queen, when the position 1/2 move later was the same either way.

And then there's games like this from older engines (from NM aww_rats):

I'll agree that newer engines are much better and quite possibly unbeatable by humans, but they still don't understand openings or endings, and shouldn't be used to measure which ones are best. In the KID (and benoni as said before) the engine algorithms say that white is winning because the algorithm likes the extra space. Then, when black kingside pawn storms (also mentioned above), it suddenly changes its mind and realizes that black has good chances too. The algorithms also like gaining tempi more then they should, they will undevelop pieces to gain a few moves. This, sometimes they completely misevaluate positions.