An n8n news brief scores every article for $0.00004 before any expensive model runs. Copy that order in your automations

The interesting part of this n8n workflow is not the news. It is the order of the steps.

What happened

n8n published a Verified Node Spotlight, written by Tyler Eastman of You.com, for a workflow that sends a morning AI news brief to Slack. A schedule trigger fires every weekday at 8am. The You.com node searches the past day of news for each topic, and articles seen in an earlier run are skipped.

Every new article then goes to Jev, a model from TypeSafe that does not write text. It answers three typed questions in one call: does this report a real development, is it about the topic, and how much does it matter to this team. The post puts the cost at about $0.00004 per article. Only articles above a materiality threshold continue. They are grouped by topic so six outlets on one launch become one item, and You.com Research writes a short brief with linked citations.

The post says a full morning costs between one and four cents. A topic with nothing material gets no post at all. The template is free to import, and Jev is in early access.

My take

This is the pattern I want in more client automations: search wide, filter cheaply, and run the expensive step last on what survived.

Most lead intake, inbox and ticket flows I audit do the opposite. Every record goes through the biggest model with one vague prompt like "is this worth a reply?". You pay for all of it, and when the output drifts you cannot tell why.

Three ideas worth stealing:

1. Ask separate small questions. "Is this a real buyer?", "Is it in our service area?" and "How urgent is it?" can each be tuned on their own. One combined score hides which judgment went wrong.
2. Keep the threshold in a config node, not buried in a prompt. Changing one number beats rewriting instructions every week.
3. Let quiet days stay quiet. A daily alert that fires when nothing happened trains people to ignore it.

Add deduplication on top, so a story or a lead that shows up three days running only gets handled once. None of this needs new tools. It needs someone to decide what gets checked before the money is spent.