In my experience using Claude code for Go and Java code, I've seen little advantage for one language over the other in the LLM agentic context. I did a little experimentation with Zig which was less successful. Presumably to do with the relative lack of documentation and still being a somewhat moving target. That said, I have had Go concurrency code written with weaker models prove to be buggy, which shows up rapidly when reviewing with stronger models.