Building in public: how our AI agents run a content factory

Description and use

This is a weekly diary of an AI-agent-run content factory: agents draft articles, qualify leads, build SEO pages and read analytics, while humans set direction and approve what ships. Failures are documented openly, and metrics are kept in a separate system-generated block.

Problem

How to run a content production line where AI agents draft articles, qualify leads, build SEO pages and read analytics, while humans set direction and approve what ships.

Outcome

19 AI agents running; 82 pieces published (20 RU, 30 EN, 32 Altezza); 4 generated programmatically; 363 leads in the system. The system is documented week by week including failures.

Workflow

• One agent drafts articles from a brief and keyword plan.
• Another agent qualifies inbound requests and prepares replies for human review.
• A generator builds structured SEO pages from verified data.
• An agent reads results and flags what to change next week.
• People choose topics, set tone, approve what ships and talk to clients.
• The system adds weekly metrics as a separate block so prose stays honest.

Evidence

Our AI agents draft articles, qualify leads, build SEO pages and read analytics, while people set direction and approve what ships.

Source

Narrow roles beat one clever generalist. When something breaks, you know where to look.

Source

We add validation rules, ban fabricated data, and make every agent say 'no data' instead of guessing. That last rule saves us the most pain.

Source

Yes, if it starts small. Automate one routine task, add human review, and expand only after it works reliably.

Source