Writing compiler is the trivial part. Weeks, without any LMMs.
Designing a language is the most complex part. Not something delegatable to LLMs (unless you want a shitty language). Takes years. One needs to study most of existing languages - and to the most of proficiency. Think deeply on computer science theory, even advance it in some spaces. Heavily invest into studying abstract mathematics like category theory, type theory, intuitionistic logic. Re-iterate hundreds of time. This just can’t be speeded up.
Then, once you have a compiler, you have to build a standard library and core ecosystem of libraries. Again years and not delegatable to LLMs (there is nothing written in the new language to train them on).
So decade(s), even in the age of shitty LLMs we are living in.
The failure of Rust language to ship with a proper concurrency is a language killer. I have fought it for 4 years with @cloudhead by applying reactor pattern in RGB, Cyphernet, @radicle - but now I am giving up.
Rust is for the business logic, which must be separated from a scalable I/O (client or DB connectivity). For backends, this business logic has to be called via ZMQ-styled MQ from a runtime written in some other language with a proper native concurrency (Go or Elixir), not a retrofitted/backported one (like in Java, C#, Python etc).
A niche for alternative language which fixes the gap between native concurrency and low-level system safe code is there. That is one of @CationLang goals, but it will take decade(s) to fulfill.
I have spent a lot of my time over the last two years designing a programming language which allows to (relatively easy) write a correct by construction code, i.e a code which formally verifies itself and does a lot of logic at compile time.
The language is called cation @CationLang, which is a feneric programming language. Its dialect Contractum is made for smart contracts running @UVioletAI (RGB successor and superset).
You may think of it as Lean 4, but able to do what you usually do with Zig.
The new age of vulnerability discoveries highlights the importance of such language. Human programming is still at the amateur stage, and we need a new generation of tools to write software which is provably safe.