@marcoarment The idea that everyone turning to LLMs to generate their code means the code it generates doesn't matter is silly. Will there be wide swaths of slop where it doesn't matter? Sure, but for the places where quality matters (life safety, mission critical, in theory places like Apple) there will still be a need to review, tweak, and smooth the source code that an engineer first generates with an LLM tool

@marcoarment I suspect we’ll have a new round of languages that are designed to be used by LLMs more efficiently, but still readable/understandable by humans.

@marcoarment i'm thinking: probably not. all of our existing languages have been for human understanding.

vibe coding still writes languages we've been using for years. what if we throw away the part where you get to audit it for the sake of a more efficient language?

that sounds crazy and reckless and totally on point witb 2026

@nebhale @marcoarment I'm curious how this LLM will be trained to use an LLM-friendly language.

@marcoarment Probably. And I actually think it has very little to do with LLMs. I feel like we’ve reached peak programming language.

@marcoarment all these real engineers with free time?

You know they're making languages.

@emc @marcoarment actually I think this is even less likely. One reason LLMs are *so* good at writing code is because there is an insane amount of material for them to train on

What about current languages is inefficient for LLMs? What aspects of LLMs would you take advantage of with a new language?

How would you train all the edge cases and best practices if there's not a huge number of humans out there generating the content LLMs have trained on so far?

@marcoarment Haskell proves that people will make programming languages that no one wants or needs, so I say no.

@marcoarment
Definitely no, but "You've made your last major programming language, {Villain_Name}!" makes for a great line.

@marcoarment Hell no. There’s usually a new version of Swift to learn every year.

@marcoarment I seriously doubt it. I’m not sure we’ll ever see another C++, but another Python? Absolutely.

@troldann we’re already using LLMs to generate the training data that’s passed in to train the frontier models. What if we asked that first LLM to generate a language that’d be easier for it to work with in the future and passed that in as training data? It’s basically a compiler bootstrapping pattern for LLMs.

@marcoarment I asked someone who worked on the first GitHub copilot if we need “a programming language for llms” his answer was that the existing ones are already perfect for it. So unlikely we will see another big one —

@marcoarment I should think that large scale concurrency has been solidly, intuitively, easily, and safely handled that there should not be any more languages needed.

@marcoarment I feel like some variant of Lisp will end up being a universal language when machines are doing most of the raw writing and reading. But maybe that’s just me (as a former professional lisp coder)

@jsonbecker @marcoarment There's something to this. I'm working on an app that has a little purpose-built scripting language and LLMs can work really well with that when provided with a reference markdown file of the syntax and supported functions, plus a CLI to evaluate the code. It's pretty remarkable seeing that in action.

@marcoarment We as in humans? Yes. I bet we see a LLM-created (or inspired) programming language. The key is data. Reasoning will be key here.

@marcoarment My totally unscientific take: yes for the „user space“ (App Development) no for kernel space.

@marcoarment Yes COBOL was it

@marcoarment propably. End of an era I guess.

I get that LLMs are the futures. But it has just killed my interest in tech/programming and I’m actively switching to a different career.