This is not a you're-holding-the-llm-wrong problem though, AI tools are simply not capable of creating mental models for problem solving. Sounds like you're tired of hearing that LLMs are not a silver bullet.

Exactly, there's things you aren't shouldn't do with an llm. But generating helm charts, configs, action workflows, building specs and then implementing based on them? Simply a no-brainer.