Cheap Chinese single-board computers made specifically for inference is the missing puzzle piece. (No, GPU's and especially Nvidia is not that.)

Also the current crop of AI agents are just utter crap. But that's a skill issue of the people coding them, expect actual advances here soon.

The smaller models are becoming even more capable now. Add that with a suite of tools and integrations and you can do most of what you do online within the infra at home.