· XingAI Invest AI
Treating LLM Output Like Cache Rows (Planned): Worker-Owned Inference
This post documents a decision we accepted, not something already running in production. It belongs in the same story as our CQRS market cache: derived data should have one writer.