To me, the most interesting part is the architecture of the engine. How you encode a position or a move, how you design the move generator, etc.

Something as simple as encoding the 64-square board in a 64-bit number leads to a rabbit hole that can last years.

After that point, the "AI" part is already solved and you just need to implement it (alpha-beta + all the heuristics).

Most of the recent chess engines simply implement a neural network (by copying Stockfish) and suddenly are at 3000 elo with no effort! Such a waste.