Use Cases Features Compare Atrium Blog GitHub
Talk to Us
Recurring Content

The Weekly Digest That Writes and Publishes Itself

Pinned sources, checked on a schedule, filtered and summarised by an analyst, posted to your channel. The recurring publication that quietly costs an analyst-day a week — running without one.

All features

Runs on a schedule, not on someone remembering

A cron schedule fires the workflow. If a run is still going when the next one is due, it skips rather than piling up; a schedule that keeps failing disables itself instead of firing into the void every week.

Filtering is a separate job from gathering

A scout checks the pinned sources; an analyst decides what is actually worth your readers' attention. Most of the value of a digest is what it leaves out.

Publishes where your audience already is

Posts straight to the target channel. The same run can notify your team with the outcome so someone knows it went out — and what went out.

You stay in the loop by exception

Approve before publishing, or let it publish and read the notification. Both are configuration, not a rewrite.

The job

“Every Monday somebody has to check the same fifteen sources, work out what changed, write it up, and post it. It is not hard. It is just relentless.”

Recurring publication is the clearest case of work that is valuable, repetitive, and quietly expensive. Nobody plans to spend an analyst-day a week on it. It happens because the alternative is to stop, and stopping is visible.

The trap in automating it is that a digest is not a summary of everything — it is a judgement about what mattered. Naive automation produces a wall of links nobody reads, which is worse than nothing because it costs credibility too.

What Forge does

Three jobs, deliberately split so that judgement is a step of its own:

Scheduleinputfires weekly
ScoutResearchchecks the pinned sources
AnalystAnalystfilters, ranks, summarises
PublisherCommunityformats and posts to the channel
Publishedoutput

The scout sweeps the sources you pinned — releases, funding events, regulatory changes, whatever the beat is.

The analyst does the part that makes it worth reading: drops the noise, ranks what is left, and writes it in the register your audience expects.

The publisher formats it for the destination and posts it.

Because these are separate steps with a validator between them, a weak week is visible: the analyst can report “nothing significant happened” instead of padding the issue — which is what a human editor would do and what naive automation never does.

Evidence

We run this shape ourselves and for partners. One live example, described without the client’s name because the channel is under NDA:

  • Weekly cadence. The scout checks the pinned industry sources on schedule.
  • The analyst filters and summarises into the established format.
  • The publisher posts to the target Telegram channel — the publication itself, not a draft in someone’s inbox.

Internally we run the same pattern for our own engineering digest, on a schedule an operator can change by messaging the bot in plain language — no dashboard, no redeploy.

What you control

  • The sources. Pinned explicitly. The scout does not wander.
  • The cadence. A cron schedule you can change conversationally; overlapping runs skip rather than stack.
  • Publish vs. approve. Either post directly, or hold for a human press of a button. Same workflow, different setting.
  • The record. Every issue keeps its handoffs, so “why did this end up in the digest?” has an answer.

Under the hood

Have a publication that eats an analyst-day a week?

Bring the source list and one issue you consider a good example — that is enough to shape the workflow.

Talk to Us All Use Cases