Podcast Agent — Deployment
The show is made up of a few small pieces that each do one job, plus a public-facing piece covered fully on Podcast Agent — Storage & Feed. This page is the map of how they fit together.
The pieces
-
The podcast agent — the one component that actually produces episodes, for all three shows. It runs continuously to collect stories, runs once a day per show to publish an episode, and can also be triggered on demand — either through its own HTTP surface (any show) or, for the tech show specifically, through a natural-language chat/A2A request from another agent in the fleet. Nothing outside the cluster can reach it directly.
-
The public podcast feeds — a small, separate piece that just serves each show’s finished episodes and RSS feed to the public, with no login required.
-
The operator dashboard — another separate, read-only piece giving a browser view over what’s been collected and published across all three shows, for internal use only.
Splitting things this way keeps blast radii small: the piece that’s reachable from the public internet (the feeds) and the piece reachable by a logged-in operator (the dashboard) can only ever read — neither can touch episode generation or the archive itself. Only the podcast agent, which nothing outside the fleet can reach directly, is ever allowed to write.
All three shows run from this same Deployment and the same set of scheduled jobs (one set per show) rather than three separate services — they were three separate repos/deployments until 2026-08-26, consolidated once the underlying code became one multi-channel service. See Podcast Agent — Overview for why.
Read next
See Podcast Agent — Storage & Feed for what’s stored and how the public feeds and dashboard read from it, and Internal Tooling Behind SSO for what the dashboard actually shows.