THE FORMAT

Every episode has the same six segments.

The recurring shape is the product. You should know exactly where to find the outcome, the architecture, and the code — no matter whose loop it is.

The six segments
#SegmentWhat goes in itRule
1Cold openThe outcome, in one concrete scene. A number, a screenshot, a moment.No setup, no context. Hook first.
2The problemWhat the human was doing manually, and why it didn't scale.Must name the real cost: time, money, missed items.
3The loopWhat runs on a schedule, what the agent decides, what deterministic code executes. One diagram or pseudocode block.Separate agent judgment from script execution explicitly — this split is what makes loops reliable.
4What broke2–4 real failure incidents and the fix that came out of each.Required. An episode with no failures is marketing, not a case study.
5The numbersRun duration, volume processed, hit rate, cost — each number cites its source.Unverifiable claims get cut or marked ⚠️ unverified. Show the receipts.
6Take it homeThe reusable artifact: code, prompts, or design rules — enough to rebuild the loop in your own domain.Must be runnable or directly adaptable, not a concept list.
The submission bar

A case qualifies for an episode if it has all four:

  • A loop that ran unattended on a schedule for two weeks or more — not a one-shot agent demo.
  • At least one measured outcome the author can show evidence for.
  • At least one failure story, with the fix.
  • Willingness to share the architecture. Full code is optional; a sketch of the prompts or rules is required.
Why the bar is strict: communities reward verifiable specifics and tear apart vague claims. Every "agents did amazing things" post without receipts makes the next one harder to believe. This site only adds to the first pile.
Voice

Documentary, not tutorial. Past tense, real names for tools, real numbers. The reader is a developer deciding whether this pattern is worth copying — respect their time.