This might handle scripts as you described, but just wait until it needs the context of random bits of a tens-of-millions-of-lines monorepo plus knowledge of custom infrastructure that isn’t documented anywhere and–oh wait, we can’t actually let this LLM-as-a-service read our code because X and Y compliance/security/legal/etc, even if we ran it on-prem.
The robots aren’t coming for you so soon, don’t worry.