Take a look at K8s for example, lots and lots of YAML configurations, which was used for human readability, but with LLMs do we still need them?
Wouldn't it be easier with LLMs to say: expose this service with 10 pods and enable only internal routing - which then updates couple lines of Python code (pulumi, aws cdk and etc,.) instead of generating 5 yaml files each with at least 40 lines of YAML and also try to reuse same attributes/labels/tags