The longer I think about the impact of AI on software development, the more I see parallels between "pets" and "cattle" types of software projects. Or more aptly, fine china and disposable styrofoam plates. LLM-based development allows large-scale production of essentially disposable software. Which is fine for many business cases - software is a cost for them. Even with software as products, it makes sense to optimize production costs.
But open-source software is far more about the process than the product. And therefore longevity of resulting software can be expected to be far greater. Foundations of our software stack - libraries, source control operating systems, compilers, databases and numerous other pieces require people capable and willing to do a long-term care of the software - hence, the community that need to last decades. Linux and PostgreSQL is 35 years old, git is 21, OpenSSL 28.
I believe human-only software development will remain for the foreseeable future, despite the growth of LLM-based ones, just like mass-produced grocery bread didn't displace the artisanal bakeries.
Why can't LLMs be used for long-term care of a software project?
This seems to be more on the tool-wielder than the tool.
Linux itself is receiving LLM-assisted code (a quick search saying 20% at the moment, although I don't know how accurate that is)