I think in principle LLMs are no different from other lowercase-a abstractions that have substantially boosted productivity while lowering cost, from compilers

I think in principle LLMs are no different from other lowercase-a abstractions that have substantially boosted productivity while lowering cost, from compilers to languages to libraries to protocols to widespread capabilities like payments and cloud services and edge compute and more and more. There is so much more software that can be written, and may be rewritten, abstract machines that can be built and rebuilt, across domains of hardware and software, that become enabled by this new intelligence-as-a-service capability. I think juniors are a significant audience for LLM code production because they provide tremendous leverage for making new things. For more experienced folk, there are lots of choices that resemble prior waves of adoption of new state of the art tools/techniques. And as it always goes, adoption of those in legacy environments is going to go more slowly, while disruption of legacy products and services that have a cost profile may occur more frequently as new economics for building and then operating something intelligent start to appear.