Quality, cost, and latency. You can usually have two. Here is the framework I use to pick which two on purpose, with a voice assistant, an RSS scorer, and a Frigate box as the worked examples.
The gap between an idea and a working product used to be an engineering team. Here's the whole loop I run now: spec first, backlog in Gitea, Claude Code doing the building, and me making the calls that actually matter. Including the two features I shipped and then threw out.
I planned Ceres with seven specialist agents and cut half of them in the first week. The surviving ones still broke, just not the way I expected. They gave me good answers and then quietly assumed somebody acted on them. What that taught me about restraint, state, and where the real work in a multi-agent system actually lives.
I have wanted an Alexa that doesn't phone home for years. Every commercial voice assistant is a microphone in your house wired to someone else's computer, and the whole point of a homelab is that my stuff runs on my…