I’m aware that it is possible to write code in garbage collected languages. Not really sure what you were addressing there.
caleb-bb
Schultzer
Your assumption that certain workload is prohibitively slow in garbage collected languages is wrong.
If you wanted say that memory management is harder to control in those languages then you would be right. It’s a trade off. Writing high performance code is a skill and can be done in any language. If performance really mattered to anyone they would write assembly or machine code.
As C can be prohibitively slow.
0
caleb-bb
So performance does not matter to anyone… and if it did… then they’d write in machine code by hand.
Gotcha.
0
Schultzer
josevalim
Creator of Elixir
Sure, if you include performance, then the equation can benefit Rust. Just as including concurrency/distribution can make it benefit Erlang/Elixir.
2
plcholder
Wow, I must say, this is the first time I’ve heard someone bring up combinatorial explosion in the context of AI agents. I’d never really thought about it that way, maybe we will never reach the “AI write me a perfect program and dont make any mistakes” promise land 
you have me rethinking that rust will be the only language left if LLms become even better and cheaper
0
Schultzer
Rust suffer the same problem as it is Turing complete as well, it might even be worse as you can completely sidestep the borrow checker with one keyword in the language.
I never said that we wont ever arrive to a point where AI is not good enough, but randomly generating tokens over an infinite plane is a problem. And we already know why it is. So what is a strength for human creativity is an achilles heel for a token generating machine as it’s hard to consistently generate good answers efficiently. The age old solution to this problem was always constraint languages.
1
caleb-bb
I think Rust won’t be the only language left. It is likely to approach dominance as a systems language, however, which it was already doing prior to LLMs.
If you use an LLM to generate Rust and wind up with unsafe Rust, then you don’t know how to use an LLM or write Rust or neither, full stop.
Agreed. The obvious benefits of the BEAM lie in orchestration/deployment but I think there’s a lot to be said for data ingestion. Any place where you’re ingesting a lot of data (e.g. RAG pipelines) there’s a potential application for OTP, as you well know.
1
AstonJ
I hadn’t seen that, tho didn’t fancy sitting through an hour of it so got Google to summarise:
The TLDR of David Heinemeier Hansson’s (DHH) opening keynote at Rails World 2026 is that the era of writing code by hand is officially over, and software developers must transition into high-level creators who guide AI agents with total optimism.
Core Takeaways
- Pencils Down on Manual Code: 37signals has completely stopped writing code by hand as a normal business practice (20:55). Manual coding is now treated like an error state or a bug (21:07).
- English as the New Programming Language: DHH declared English as his favorite programming language (38:12). He argued it has surpassed Ruby because it allows developers to program through intent and outcomes rather than implementation details (38:29).
- Reengineering HEY & Native Apps: Instead of maintaining traditional web or hybrid apps, 37signals used agents to quickly build native front-ends across platforms (26:56). Concurrently, AI rewritten back-ends in Rust reduced CPU usage by 99%, making it possible to serve peak traffic on a single Raspberry Pi (30:49).
- Omarchy & One-Shot Applications: DHH showcased rapid operating system innovation with his project Omarchy, which achieved an installation time of just 9 seconds in the lab (48:10). He also generated lightweight, highly optimized “one-shot” apps within minutes, such as a markdown presentation software called Hype that fits into a half-megabyte binary (51:36).
- Total Optimism over “P(doom)”: Addressing industry anxieties about AI replacing programmers, DHH looked to history to note that paradigm shifts universally unlock abundance (58:34). Citing Jevons’ paradox and how ATMs actually increased the number of bank tellers, he urged developers to choose “total optimism” and embrace their new roles as high-fidelity creators (41:12).
What was more interesting to me tho was the interview with Matz afterwards:
Back to the main topic of this thread, I think it’s great that José (and others) are thinking about the future of programming languages. This in particular is good to hear:
In the past, I have said that creating great tools for humans also leads to great tools for agents.
For the near future, while humans are still involved (whether programming directly or having to direct/check what AI tools write) then I reckon it’ll be beneficial languages remaining enjoyable and human-centric - the more enjoyable and intuitive they are, the more likely we might choose them.
At some point there will come a time when we won’t need to write (or manually check) code at all - but - there is also almost certainty that AI companies will be at the stage where they’ll create their own AI-optimised languages and runtimes. What then? Where will the languages we know and love be then? DHH, in the post show panel/interview with Matz above, basically said he doesn’t care about programming languages anymore - he’s using Rust today, but won’t hesitate to drop it for something better tomorrow. So he doesn’t really care whether Ruby lives or dies because he no longer ‘needs’ it.
So we have to ask ourselves - do we care?
I saw a video a few days ago that said keep programming because you won’t be able to use AI services to create ‘illegal’ apps - and that does give you pause for thought. When might we not want to be beholden to the whims of AI companies or certain govts? Will we always be? Will local LLMs and AI tools be good enough? If not, what will that mean to us? Will we still need a human-centric language then?
Personally, as I’ve said since pretty much the beginning, AI is still in its infancy, so imagine where it will be 5, 10 or even 50 years from now - the only thing that concerns me - is that those currently at the helm just can’t be trusted. And this is why I personally hope language and framework creators will continue to be vigilant and to make their languages and frameworks intuitive and human-centric… at least for the foreseeable future and until a time where we have full trust in our new found friends and their controllers 
6
derek-zhou
It is very sad to learn that the father of Rail is abandoning Rail, announcing at the most prominent Rail conference, no less. Fortunately for me, I didn’t invest too much on Rail.
I know I will still need it because I don’t write code for a living; I write code for fun. Plus, I don’t find bossing someone around, who doesn’t fight back, any fun.
4