No new connection initiating an additional concept or function is done without reference to the existing architecture.
Imagine the network of neurons as a game of pieces of wood that you must stack on top of each other. You have an unlimited number of pieces. The more a piece is placed at a high height, the more points it gives you. The pieces do not all have the same shape, some assemble very stably, others fall at the first jolt. You do not have to add them randomly. If you have trouble placing a weird shaped piece, you can leave it on the floor, but it does not earn you points. Or you can readjust the arrangement of the highest floors of the stack so that the bizarre piece coincides and occupies one of the peaks.
The wooden pieces are the concepts built by the neural networks and the height of the pile is your intelligence for each of the subjects analyzed by the brain. You can only add a higher concept to each stack if it respects at least part of the architecture that underlies it. Incompatible concepts are rejected, fall to the floor. Those who are difficult to hold at the top but bring great benefits tend to maintain themselves (celebrity brought by the stimulation of reward circuits) and have a retroactive effect on the underlying structure, rearranging the sub-concepts to become concordant. This is how a notion that is difficult to swallow but rewarding becomes better and better understood as we use it.
The dialogue between the new connections and the old ones is therefore permanent and bidirectional. There is no definitive structure, no fixed function, other than by its regular and profitable use (in terms of coherence of concepts between them and not according to any “truth”).
Finally, for my analogy with the wooden pieces to be accurate, you must eliminate the player. Indeed the pieces are assembled alone. This makes it easier to understand why connections are not random. Many options are possible, but it is the efficiency of the new connections created that establishes their permanence. There is no programmer who would be satisfied with the result, based on a pre-established goal of the brain’s destiny. The ideas in question are part of the self-organization of concepts.
Reason why you can not know, even if your parents apprentice-programmers had specific expectations about it, what person will you be in 1 year, in 10 years, in 50 years ..