It's only OpenAI which doesn't have an in-house modern runtime

https://openai.com/index/openai-to-acquire-astral/

They should just go all in with Wasm.

~/.local/bin/codex --dangerously-bypass-approvals-and-sandbox -c 'model="internal-gpt-7-turbo-blackhole"' -c 'model_reasoning_effort="max"' -c 'plan_mode_reasoning_effort="max"'

create a plan to implement a javascript runtime similar to {deno,bun,node.js,whatever} in {Rust,C++,ASM,Go,Fortran} language, use subagents @ max effort, keep building until you're done, follow instructions according to INSTRUCTIONS.md

Wait a day and you're done.