I think there's a really stupid point I want to make.
Follow me on this.
A few things have happened in the world of coding and programming languages in the debate about agentic coding.
The debate on legibility and code review in the age of agentic coding is alive and well.
Some people are in the "humans shouldn't manually review or write code anymore" camp.
And that's fine.
I go in and out of that one myself.
But if you think about the implications of that, there's a new set of conceits that we should probably think about that we've never considered before.
Because if humans aren't reviewing code anymore, the code no longer needs to be legible.
And if the code doesn't need to be legible... human readable programming languages in general, are no longer necessary.
Two constraints now come into play:
1. What's the cheapest possible way to generate this new non-human legible code?
And
2. What pre-computed priors in the LLM can we easily exploit to reach those two previous objectives.
The answer is simple, don't you think?
It's pretty obvious.
All we need is a composable Brainfuck compiler with a standard library.
If done correctly, it can outperform Rust, because it's a much simpler language.
And the machine already knows how to work with it.
Think about the massive token savings you'll see when your code is nothing but incoherent punctuation.
You're welcome.
No, I'm seriously going to build this.