Well, I see LLM coding capabilities as a great enabler for people who have some codeing-like skills or needs, but were not sufficiently skilled to do something more complicated. Think of people who are good at Excel, who use statistical tools like SAS, SPSS, other analytical software. Now they can ask LLM to create a Pandas/SAS lang script and do much more advanced stuff.

People who were in the marketing data analysis (like sentiment analysis) - 5 minutes and they have a code that uses Hugging Face model suited for sentiment analysis, zero-shot classification, etc. No need to pay for expensive online services or expensive NLP software. It's here for free or $20 a month.

Still, it does not mean you will be able to code database engine with LLM, application server, rewrite Django in Rust, etc. So software engineers still will be needed to do ambitious, complicated stuff.

So, I kind of see it backwards, real skills, like knowing algorithms, understanding performance (including hardware stuff like processor caches, etc.) will become needed, as other, simpler jobs that needed only a "coding monkey" will be gone.

We no longer need to dig ditches manually, we have machines for that, but the purpose of the ditches is still planned by man.