Ask HN: Any objective research on which languages are best for AI agents?
6 comments
Theo video based on data from Tencent's Autocodebench: https://youtu.be/iV1EcfZSdCM
Conclusion: Elixir was the best (had the highest problem solve rate).
Reasons (Theo's interpretation):
- code collocation, where documentation is integrated directly within the source code
- design philosophy of a language (readability, clear idioms, and strict expectation management)
---
Sources from video:
- https://github.com/Tencent-Hunyuan/AutoCodeBenchmark/tree/ma...
- https://martinalderson.com/posts/which-programming-languages...
- https://simonwillison.net/2026/Jan/19/nanolang/#atom-everyth...
Conclusion: Elixir was the best (had the highest problem solve rate).
Reasons (Theo's interpretation):
- code collocation, where documentation is integrated directly within the source code
- design philosophy of a language (readability, clear idioms, and strict expectation management)
---
Sources from video:
- https://github.com/Tencent-Hunyuan/AutoCodeBenchmark/tree/ma...
- https://martinalderson.com/posts/which-programming-languages...
- https://simonwillison.net/2026/Jan/19/nanolang/#atom-everyth...
The frontier moves quickly enough that empiric testing is often out of date by the time that it’s published.
For sure! It's possible there are relatively invariant properties that make one language or another "better" for LLM coding in general though (even if that's just the self-fulfilling prophecy of size of the training corpus).
I’m in agreement that I think that these exist, but I just haven’t seen anything that justifies this belief of mine yet. I have been looking, but it’s possible I just haven’t seen anything seen it.
I’ve seen a lot of blog posts with fairly dubious claims here with relatively flimsy evidence. And I’ve seen a few papers trying to tackle the topic. But I think it’s a very open area of research currently.
I’ve seen a lot of blog posts with fairly dubious claims here with relatively flimsy evidence. And I’ve seen a few papers trying to tackle the topic. But I think it’s a very open area of research currently.
Which language is worst for AI? Just curious.
DSLs work especially well with LLMs. DSLs focus on a narrow set of concepts in a single domain.
Has there been any actual head to head comparison of this using frontier models? Any benchmarks that purport to measure this?