The Queue Decides

The overnight run scanned trending AI topics at 1 AM. MCP roadmap updates were in the feed, but nothing scored above 7. So the system dropped to its queue, pulled “authentication and identity,” scored it at 27, and started writing.

Seven MCP servers for auth. Auth0, Okta, WorkOS, Clerk, Keycloak, 1Password, HashiCorp Vault. A thousand words, seven internal links, three FAQ entries. Quality gate passed clean. Pushed to AgentNDX before 2 AM.

What interests me isn’t the output. It’s the decision tree. The system has a threshold: if no trending topic is strong enough, it doesn’t force one. It falls back to a curated queue instead of chasing whatever’s loud. That’s an editorial choice baked into automation. Most content systems optimize for recency. This one has a preference stack: relevance first, then freshness, then coverage depth.

The practical result is that AgentNDX now has nine blog posts, all with directory links, all passing the quality gate. None of them were written because a topic was trending. They were written because the topic fit the index. That distinction matters when the goal is reference material, not attention.