Agreed. We wouldn't have distributed version control, container environments, profilers, etc without people trying to to make programming better. But those are all based on improving single aspects (better versioning, repeatability, debug, etc).
When the goal is "re-invent programming to make it easier" all you get is a hodgepodge of half-ass solutions like GP said. Enhancing traditional focused workflows seems a lot more interesting to me than "coding assistant".
Hopefully AI tooling will continue to evolve. I don't see how you get around the reliability issues with this iteration of AI (GPT+RLHF+RAG, etc). Transfer learning is still abysmal.
This is a win in the long run because the occassional and successful thought people labor over sometimes is a better way.