Two-stage generation, and who is waiting for the answer
what decides which transport an LLM call gets is who is waiting for the answer. Everything else in this post follows from that one line. The system is small and unglamorous: an ops utility in the mening repo, cmd/socialthread, that drafts one social post a day for a language-learning app. It runs on my machine, it talks to nobody’s users, and a human presses publish. That combination is what makes it a clean place to think about transports and about verification, because both questions show up in a form small enough to hold in your head. ...