Every operator has the same Monday problem: the industry moved all week, the information exists, and it is spread across 150 items in newsletters, Hacker News, and blogs that nobody has time to read. This agent reads all of it before dawn and hands over the only part that matters: one brief, filtered by what its owner actually cares about.
What does this agent do?
Every Monday at dawn it pulls every source on a hand-picked feed list — newsletters, filtered Hacker News, industry blogs, all free RSS. It throws away everything it has already seen, runs what is left through a written relevance filter that knows the owner's business, and delivers a single ranked brief to a chat channel. The whole thing reads in five minutes on a phone.
The output feeds more than a reading habit: the same signal store supplies this site's content engine and the weekly strategy digest, so one Monday run powers several downstream consumers.
How does the Monday run work?
- Fetch. A zero-dependency fetcher pulls every feed on the source list. One dead feed never kills the run — failures are named at the bottom of the brief instead of silently vanishing.
- Dedupe. Every item is checked against a persistent seen-list of every link the agent has ever surfaced. Only genuinely new items continue. Nothing ever repeats.
- Filter. The survivors pass through a written filter file: who the owner is, what they are building, what matters (deals, tools, competitor moves, regulation), what to ignore, and the bar every item must clear — "would I change something I'm doing this month because of this?"
- Brief. At most ten items, ranked by relevance rather than popularity, each with a one-line "why this matters to you." The brief closes with its own audit line — items fetched, new after dedupe, kept — so the filter's judgment is inspectable every single week.
What keeps the brief honest?
The counting is code, not vibes. A recent Monday run pulled 143 items from 7 free sources, deduped them to 88 new, and distilled those to a 7-signal brief — and those numbers come from the run's own telemetry, not from memory. The filter file explicitly forbids padding: a quiet week produces a short brief, because an agent that invents importance to fill ten slots teaches its owner to stop trusting it.
Data cost is zero. Every source is keyless RSS, and the model runs on a flat-rate subscription that was already being paid for. There is no per-week marginal cost to stay informed.
Where does the human stay in charge?
This agent is read-only by design. It never posts, never replies, never acts on the news — it summarizes and stops. The judgment lives in a plain-text filter file the owner writes and tunes; editing that file is the entire steering wheel. When the brief gets something wrong, the fix is a sentence in the filter, not a code change.
What value does it produce?
Two hours of scattered weekly reading collapses into five focused minutes, and — the part nobody advertises — the owner gets comfortable missing things, because anything that matters will surface in the brief. Signal quality compounds: every week of filter tuning makes the next Monday sharper.
What could this look like for your business?
Every niche has operators drowning in feeds, and this exact agent pointed at a different source list is a weekly industry brief for your industry — or for your clients: onboarding a client is a 30-minute call to fill in their filter file and feed list, and the same machine serves each one from its own folder. We publish the complete build — fetcher, filter template, cron setup, and the client-service playbook — as a free kit, so the fastest way to evaluate this agent is to run it on your own feeds next Monday.