In the traditional start position (SP518: RNBQKBNR), we take for granted that two minor pieces must move before castling to either side is possible, and that the Queen must move before castling O-O-O is possible. In chess960, those considerations are different for each start position. In the worst case, positions like RKxxxxxR, five pieces must move before castling O-O is possible, while in RxxxxxKR five pieces must move before O-O-O is possible.
In SP518, we have guidelines like Knights before Bishops, when the best developing square for a Knight is often more obvious than the best square for a Bishop. In other chess960 positions, the best square for a Knight is not at all obvious and may depend on the opponent's initial moves.
Some time ago, in A Framework for Chess960 Opening Theory,
I noted that the different SPs can be classified and analyzed generically depending on which files the pieces are initially located. It turns out that the framework is useful in chess960 for counting the developing moves and for determining best move order.
The development of the pieces can't be taken individually. A good example is the family of positions like BxNxxxxx (or their twins xxxxxNxB), where White should be wary of Nc1-b3 (Nf1-g3 for the twins), blocking the movement of the b-Pawn and the opening of the diagonal for the Bishop on a1. If White has already played b2-b4 (g2-g4), the move Nc1-b3 (Nf1-g3) is perfectly logical. Similar considerations, usually more subtle, come into play for every chess960 opening.