How our AI agents write and publish a tech blog, and where humans step in

There is a lot of AI-written content on the internet, and most of it is easy to spot: generic intros, repeated keywords, confident claims with no source. We didn’t want our blog to read like that, but we also didn’t want to spend every evening writing.

So we built a pipeline where AI agents do the heavy lifting and humans own the judgement. Here is how a story moves from a news feed to a published article.

Step 1: Collect the signal

A research agent watches a fixed set of news feeds in cloud and AI that we chose ourselves. Fixed is the key word. An agent that searches the open web for “AI news” brings back noise; an agent watching sources we chose brings back things our readers care about.

Output: a short list of candidate stories, each with a link and a two-line summary.

Step 2: Choose what’s worth writing

Athena, our planning agent, ranks the candidates against the week’s goals and what we’ve covered recently, and proposes the topics. A person approves the final list. This takes five minutes and saves hours of writing articles nobody needed.

The rule: we only cover a story if we can add context, not just repeat the announcement.

Step 3: Draft against a style guide

Amy, our blog agent, writes the first draft. She works from a written style guide that every draft must follow:

lead with what changed and who it affects, in the first two sentences;
explain the technology in plain language before any jargon;
link to the primary source for every factual claim;
no filler intros, no keyword stuffing, no invented quotes or numbers;
end with what readers should do or watch next.

The style guide matters more than the model. Most “bad AI writing” is really a missing brief.

Step 4: Human review (never skipped)

An editor reads every draft before it goes anywhere. The review checks three things:

Accuracy. Every claim matches its linked source.
Value. The piece says something the announcement didn’t.
Voice. It sounds like us, not like a template.

Drafts that fail go back with notes, or get dropped. Nothing is published straight from an agent.

Step 5: Publish and format

Approved articles are published to our WordPress blog with a consistent structure: headline, summary, subheadings, source links and a cover image. Consistency is what makes a blog look professional over time.

Step 6: Distribute

Once an article is live, our distribution agents pick it up from the feed and adapt it: a Substack version, a newsletter section, social posts and YouTube notes. They only ever work from published, approved content, so distribution can’t introduce new claims.

What we learned

Speed is not the main win. Consistency is. The pipeline publishes on schedule even in busy weeks.
Narrow agents beat clever prompts. One agent per job is easier to improve than one agent doing everything.
The review step is the product. Readers trust a blog because someone stands behind it.

Want the same pipeline?

This workflow is part of Agent Studio, which we now offer to creators and media teams in beta. If you want a blog that publishes consistently without losing your voice, talk to us.