This is the sort of work that AI is coming for. It's low impact, low effort, low paid, and importantly it doesn't get automated (often). This means there is a lot of low hanging fruit for automation, whether that's by regular tooling or LLMs.
On the other hand, if you're a highly paid software engineer, it's likely you're already working with highly automated and abstracted systems, so there's much less opportunity for LLMs. If your deployment process is a Word doc specifying the 120 steps you have to run in bash to deploy that takes 1 eng day, that's going to be automated, but if you're git-pushing and having Kubernetes roll it out in your CI/CD flow, automation has already raised the level of abstraction that you need to work at and there are fewer opportunities.
Yes there's Copilot/etc which will reduce boilerplate, but the only way a company achieves "70% less coders" is by wholesale automation of large parts of their process that must be horribly basic already.