We automated the reading, not the voice
How BARGO checks recent primary sources, prepares a sourced article and social drafts, and keeps publication under human review.
Published · Arnis Geidmanis · 3 min read
Workflow updated on 2026-09-22. The original publication date is retained above.
Publishing more is not the same as saying something useful. We built a narrow reading workflow for BARGO: find a relevant change, check its source, explain the practical consequence and prepare a draft for a person to review.
What the reading workflow checks
The workflow checks daily for recent items from an AI newsletter and selected primary publisher feeds. It asks whether a business owner, finance head or operations head should change something about documents, approvals, company knowledge or product data. A story does not qualify simply because it mentions AI.
A recent newsletter can contain an old story. We therefore check the original source publication date and the date of the underlying event separately. Both must be within seven days for a news draft. A page refresh does not make an old incident current. If the dates are missing or ambiguous, the automated path stops rather than guessing.
The job then extracts passages from the publisher’s page. Code checks that the quoted passages actually appear in the fetched text and that the event date matches its supporting passage. Numeric claims in the article and social drafts must come from the checked evidence. Suggested applications are described as suggestions, not as results BARGO has already achieved.
These checks reduce avoidable mistakes. They do not prove that a publisher is correct, or that every sentence produced by a model is a sound interpretation. That is why source links and human review remain part of the workflow.
From a source to an article and social posts
A qualifying item becomes a blog draft with its source, event date and publication-date evidence attached. LinkedIn and X get shorter versions that link to the same BARGO article. The article links to the original publisher and separates the reported change from our analysis.
The review desk shows the article, both social drafts and whether the news is still recent enough to publish. A founder can edit or reject it. Publishing puts the approved article on the blog, creates its share card and schedules the connected social channels. The date check runs again for the intended delivery time; a draft that sat in the queue too long must be researched again.
Our own implementation notes are a separate content stream. They describe BARGO’s work and can have a longer useful life than a news item. Their original dates remain visible. They are not presented as breaking news.
What happens when a check fails
If no current, relevant item survives verification, the news job produces no draft. A failed notification does not erase a saved draft or justify creating a duplicate. The admin desk keeps the delivery status visible so a scheduled post is distinguishable from a draft or a failed delivery.
We test the workflow with disposable data and mocked responses, including old stories in new newsletters, missing dates, unsupported quotes and repeated runs. Those tests exercise the rules without posting to a real account. Actual source quality, model output and external delivery still need review in operation.
The useful pattern is a focused task with explicit limits: evidence before writing, a person before publication, and a record of what happened. That is the standard we want our own marketing workflow to meet.