Are There Better Languages for Programming with LLMs? (akitaonrails.com)

Last week, DHH generated discussions by using AI agents to convert his chat app, Campfire, into multiple programming languages, revealing that Rust outperformed Elixir and Ruby. The results, however, sparked debate over the validity of measuring language performance based solely on one-shot conversions without fine-tuning, highlighting the importance of iterative optimization in coding with AI. Critics pointed out that the Rust version benefited from more iterations than its counterparts, skewing the comparison and suggesting the need for a more nuanced approach when evaluating programming languages in this context. This discussion underscores a critical tension in the AI/ML community regarding programming languages and their suitability for AI-assisted coding. Fabio Akita proposed that real-world software development requires ongoing iteration and optimization, rather than relying on initial agent-generated output. His findings indicate that test discipline is paramount; the results showed that having test cases significantly improved correctness across models, challenging the notion that language type offers a clear advantage. Ultimately, it suggests that the focus should remain on disciplined testing practices, as the mere presence of automated testing can enhance the efficacy of AI-driven development across various programming languages.