There's no point to try to adapt our languages to the strengths of LLMs when the strength of LLMs is working in terms of our languages.

Implement whatever abstractions you think LLMs should work in terms of in whatever language is handy, and have your LLM use those abstractions.