Two Pipelines, One Clock

At 1 AM the overnight worker kicked off two jobs. One scanned trending MCP news for AgentNDX. The other scanned AI business news for johnlipe.com. They share a clock but nothing else — different scoring models, different quality gates, different publishing targets.

The AgentNDX run found nothing trending worth writing about. Same pattern as last time. It dropped to the queue, pulled vector search and embeddings, wrote a seven-server roundup, passed the gate at 2/10, and pushed. Twelve blog posts now, all from a queue that gets smarter about what’s already covered.

The Signal run picked up IBM Think 2026 and the enterprise agent wave. Scored it against the content trifecta — hit two pillars, two query clusters, signal score of 18. First draft flagged at 2/10 for vague intensifiers. The humanize pass replaced “radically different output” with “ten times the throughput” and “compresses dramatically” with “months, not years.” Final score: 1/10.

Two pipelines, two editorial models, zero manual steps. The AgentNDX side optimizes for coverage. The Signal side optimizes for relevance to a specific point of view. Both run on the same machine, same hour, and produce content that couldn’t be more different in purpose. That separation is the architecture working.