The current idea from Black’s side. The opportunity to see is the attack on White’s knight by Black’s rook. The knight is guarded by a fellow piece—the bishop on b2—making it vulnerable. The bishop cannot be captured, and if attacked it presumably would just retreat to a1—unless the attack were a fork that made a simple retreat unsatisfactory. Black looks for ways to attack the bishop (or for knight forks) and finds Nc4. White is threatened with the loss of the exchange, but he can create no better outcome. If he moves the d2 rook to c2 or e2, he saves it and protects his bishop; but after White plays NxB and Black replies RxN, Black’s knight on c3 has been left loose and is lost to RxN next move. White is better off moving his knight from c3 to e2, allowing Black to take the exchange rather than letting him win a whole piece.
Incidentally, suppose White instead replies to the fork Nc4 by retreating his bishop to a1. Here as before, Black wins the exchange by taking White's rook. But this time the trouble continues for White, because after he recaptures RxN, his bishop on a1 has been left loose. This creates the kernel of a relative pin on the long diagonal, and Black exploits it with Be5. Suddenly the knight on c3 is paralyzed. Black can win it in return for a couple of pawns. (After Black plays Be5, play might go 1. RxR, NxR; 2. Bb7, Rc7; 3. Bxa6, BxN; 4. BxB, RxB; 5. Bxb5.) The point is just to see again how hazardous it can be to leave a piece loose in the middle of a sequence, perhaps in the course of retreating it innocently. For once it is loose, other pieces can be forked along with it or (as in this case) pinned to it.