Yeah, I realized this around may 2024 and started to rail models in deterministic workflows and tooling. LLMs are the new CPU.