I hope next up is the performance and bloat that the LLMs can try and improve.

Especially on perf side I would wager LLMs can go from meat sacks what ever works to how do I solve this with best available algorithm and architecture (that also follows some best practises).