I think it isn't all about deep learning and the things that are trendy today, to have modules that fit together the issue of shared semantics needs to be addressed, so I think there will be something that looks a bit like the old symbolic AI together with modules based on neural networks and other methods like that.
Very true. Most of it has to do with coding that involves prediction, statistical analysis, vector similarity and other experimental techniques that involve approaches outside the box of formal methods encapsulated in off-the-shelf libraries.