Do you review and approve plaintext plans in your org and ship whatever output Claude outputs that passes the CI to prod without further review? Because that's what we do for assembly.
I think the point is that's where all the big tech companies say we're heading. I can't say I endorse it, but the OP who just left it running for a month seems to like it.
You can determine and justify the reasons why generated assembly is generated because it's made by a deterministic machine. How is an LLM's output deterministic and justifiable? How can one hold anyone to account what spews out by a large language model?
That isn't what I've seen. It seems to use every language in the way idiomatic for it, or more accurately, in the way it has een that language be ised. Rust written that way isn't present in it's training corpus so it doesn't do that. I would be more concerned about it getting creative and adding something a cool rustacean might add in the porting process that you don't actually want.