I apologize since this is a quite vague question. And I am personally at an expert in these fields at all.

It seems to me that there are two main directions of the Langlands program, namely, construction of Galois representation, modularity (automorphy) theorems. However, in all that have been achieved, there is always the assumption that the "base field" is either totally real or CM. For example, the most recent Scholze's work on construction of Galois representation (as well as work of many other authors), or the recent automorphy lifting theorems by various authors.

Indeed, it seems to me all these work use the tool of "unitary Shimura varieties" a lot, which, is only available with CM base fields.

What is the difficulty in going beyond CM field? And what are the possible ways to attack them?

Thank you!