Privacy-first is becoming a procurement checkbox, not a nice-to-have. Every enterprise RFP I've seen in the last 6 months asks "does this tool send our data to a third-party LLM?" Teams that can answer "no, or only with strict tenancy isolation" are winning deals. Local/on-prem inference with smaller models is genuinely viable now for a lot of workflows — worth learning the on-device stack even if your current job is pure cloud.