I was going to interject that very sentiment. The upfront costs of running any model of value to the average consumer is so large to be basically non-existent. We need hardware to become commoditized and capable of handling LLMs at the scale necessary to enjoy these agentic workflows.