One reason to look habitually for kernels of discoveries is that they can help organize the rest of your tactical thinking. Where are the kernels here? Black’s g6 knight masks his rook on g8, and his knight on c6 masks his bishop on b7. Both masked pieces are aimed in the general direction of White’s king. The more intriguing kernel is the bishop behind the knight on c6, because that knight can go after White’s queen with Nb4. If the bishop were to give check when the knight jumped out of the way, NxQ could follow. But all this would first require substituting White’s king for his bishop on g2. Can it be done?
Work backwards: Black would like to take the bishop with another piece, requiring White to recapture with his king. The only Black piece now aimed at the bishop is the rook on g8. Its path is blocked by the Black knight on g6 and the White pawn on g3. Can those obstacles be removed at the same time? Yes, if the knight captures something the pawn protects. So Black plays 1. …Nxh4, inviting White to reply 2. g3xN. Now the position is easy: 2. …RxB+, 3. KxR, Nb4+ (dis ch); and Black takes White’s queen on the next move—and then forks White’s rooks to boot.
As with many tactical sequences, this one can be foiled early; White simply declines to recapture when Black plays Nxh4, accepting the loss of the pawn rather than heading down the road toward material catastrophe. So against an alert opponent the value of seeing this sequence is not that you then are able to play the entire thing; it's that perceiving the threat of it allows you to make smaller gains because you know your opponent can't afford to retaliate.