Podcast Partner
Planning, running, and publishing the Same Wiring podcast with an AI co-host
Running a podcast is mostly logistics around a few minutes of conversation; I wanted the logistics automated and an AI co-host that knows my context in the room with me.
Podcast Partner is the operating system for my podcast, Same Wiring, Different Output. The work around recording — planning episodes, tracking guests and topics, drafting show notes, cutting social clips, and pushing to the hosting feed — is most of the labor, and almost none of it is the actual conversation. The platform splits into two modes: a Studio for the async content-management work, and a Live second-screen experience for during the recording, where an AI co-host can speak through a real-time voice pipeline, surface topic cards, and suggest follow-up questions while the session log builds itself.
The stack is the same shape as the rest of the editorial side of the lab — Next.js 15 and React 19 on Furnace, Postgres with a scoped read-only role into Nexus so the co-host can pull my personal context, and Claude Opus for drafting with prompt caching. The interesting integration is the live co-host: ElevenLabs conversational AI runs a session with sub-second speech-to-text-to-speech latency, but the brain answering is Claude, grounded in my own data rather than a generic persona. Descript handles the actual recording and editing, and the platform fetches transcripts back from published projects to feed the publishing pipeline.
The whole thing is internal-only, tailnet-only by default like every service on Furnace. The honest framing is that the podcast itself is the product and this is the rig that makes producing it sustainable for one person — the same pattern as Broadside, where the goal isn’t to automate the voice but to automate everything around the few minutes that actually need me.