What you are probably reading is that it's seen as a mistake to reuse a graphics scene graph as the managed scene data structure for the rest of the program to work with.
For a long while, back in the stone age, memory was tight and there was only one data structure holding a level together and it served rendering and game logic. Now we have a little room to work with and performance and easier coding of more complex features means that we need and can use structures appropriate to the algorithms.
For example, a quadtree is great for graphics operations but it would be ridiculously bad if you needed a list of all active enemies.
Beyond that simple example I think that you need to really think about what you are asking and add a lot more detail (and a glossary) to the question before you'll get any replies useful to you.