Only Juniors can think that.
You can "vibe code" with Rust? And who is doing the reviews?
Verifying requisites, performance, security?
You must know the language very well to have a senior level.
Agreed. There's no way someone can vibe code production-quality code today...
Interestingly as AI models are becoming "more competent" I'm finding more and more issues with AI generated code in the project I work on...
Whenever AI is used by a more junior dev (or a senior dev who simply can't be assed) you always find strange patterns which a senior would never have done...
Typically the code works, but there might be subtle security issues or just unusual coding patterns where it appears an LLM has written slop, and instead of taking a step back and reconsidering its approach when errors crop up, LLMs tend to just add layers of complexity to patch over its slop.
These problems obviously compound if left unchecked.
I actually prefer how things were last year when coding models were less competent because at least if a problem was hard enough they'd get nowhere. Today they're good enough to keep hacking until the slop it writes is just about working.
In regards to OPs question though, I suspect there's less point in playing around with different technologies to get some basic understanding of how they work today (LLMs can do this). But if you want to be able to guide LLMs towards good solutions and ensure the code being produced in the era of AI is good, then having engineers with a deep understanding of the technologies they're using is very important.
Interestingly as AI models are becoming "more competent" I'm finding more and more issues with AI generated code in the project I work on...
Whenever AI is used by a more junior dev (or a senior dev who simply can't be assed) you always find strange patterns which a senior would never have done...
Typically the code works, but there might be subtle security issues or just unusual coding patterns where it appears an LLM has written slop, and instead of taking a step back and reconsidering its approach when errors crop up, LLMs tend to just add layers of complexity to patch over its slop.
These problems obviously compound if left unchecked.
I actually prefer how things were last year when coding models were less competent because at least if a problem was hard enough they'd get nowhere. Today they're good enough to keep hacking until the slop it writes is just about working.
In regards to OPs question though, I suspect there's less point in playing around with different technologies to get some basic understanding of how they work today (LLMs can do this). But if you want to be able to guide LLMs towards good solutions and ensure the code being produced in the era of AI is good, then having engineers with a deep understanding of the technologies they're using is very important.