OpenAI tests LLMs on migrating COBOL banking code to C++, Java, Kotlin, Python, and Rust

Fine if LLMs just help people read COBOL. The bank still needs someone who can maintain it when the rules change. I don't want vibe-coded bank infrastructure, but I do want LLMs that can work with COBOL.

Vibe coding isn't the hard part, maintaining it is. Way more people want to write new Rust than want to maintain somebody else's vibe-coded Rust, so the old system keeps running until it's the next guy's problem.

Why would I use anything but Rust, outside of platform limits like JS or Swift/Kotlin? The agents make it work, the code doesn't look cluttered, and it's fast.

Choosing a language as a temporary decision sounds like improper planning to me. I'd rather have code that's been tested over a long period than a codebase that changes every few months.

Clojure isn't really low resource in my book. I had good results with it over a year ago.

Writing a de/serializer for an open source format with tests catching every mistake in a loop isn't insane. It would be insane to trust vibe-coded replacements with critical production data.

LLMs are pushing reverse engineering communities forward. The Shockwave/Lingo community kept at it for years, and that work is part of how the Habbo Origins client got made.

Assembly isn't on this list, and it should be. I've been building a modern assembler syntax with C-like types, and Claude codes in it fine, though I'd like real numbers. Dan Luu found LLMs underperform in assembler.