This question is a sequel to an earlier one . Just as in the openings, computers dominate endgames. Endgame tablebases provide an exhaustive analysis of endgame positions with a small number of chessmen. The undesirable side-effect is that optimal endgame moves are often extremely difficult for an endgame-expert to understand (see the last post by Bill Dubuque in the link above).
How has this phenomenon changed grandmasters' preparation against computers?
For example, they may not allow a game to continue till the endgame when playing against a computer, unlike a human opponent. Or they may avoid endgames with complex variations.
Note that the endgame allows exhaustive search, and lacks exploitable phenomena like the horizon effect .