erezT

Durable Objects for agent state is the correct move for low-latency memory. I've been building these exact patterns as a TypeScript first class memory primitive for AI agents: https://github.com/ereztdev/langmem-ts it handles the extraction/recall loop without the Python-heavy overhead of LangMem or Mem0. Pure TS, no wrappers

Most people assume podcasting works like music streaming. Someone presses play. The creator gets paid.

That’s not how it works.

Podcasts don’t get paid for listens. They get paid for ads, and only under specific conditions. The listen has to be long enough. The episode has to be new enough. The ad has to actually exist.

A million listens on an old episode can generate nothing.

What surprised me digging into this wasn’t the ad model itself, but how much discovery quietly controls income. If an episode isn’t surfaced in the short window advertisers care about, it effectively never monetizes. Chronology matters more than popularity, but most podcast apps don’t make chronology easy to reason about across shows.

From the listener side, everything looks equal. From the creator side, value decays fast and invisibly.

This creates a long tail of shows that feel popular but never really earn, while platforms still benefit from engagement.

seems like someone logged some JS right into the DOM there

I think the idea here is more into logically thinking in a critical manner, and in the last few years especially due to wikileaks, some of your worst nightmares are real AF, and then Epstein happened.

I'm slacking this to our mechanical team, they love this stuff, thanks man!

interesting