Building in public: how AI agents run our content factory

This week in numbers (real, from our system)

• 🤖 AI agents running: 19
• 📝 Content published: 72 (blog RU 18, EN 26, Altezza 28)
• ⚙️ Generated programmatically: 4
• 📥 Leads in the system: 341 (+0 in the last 7 days)

Figures as of 2026-10-14 — computed by code from the DB and files, no manual entry.

Short answer: In our content factory, AI agents draft articles, qualify sales conversations, generate SEO pages and read analytics, while people set strategy, review output and own decisions. Content automation works when each agent has a narrow job and a human checkpoint. It fails when agents run unsupervised.

We do building in public because vague claims about AI help nobody. This is the diary of how our AI agents run the content factory behind Arxitek. Real figures are added separately, so here I stick to the engineering story.

How do AI agents run a content factory?

Think assembly line, not one clever bot. One agent researches a topic and drafts an article. Another turns briefs into SEO pages. A third handles first-touch sales conversations. A fourth reads analytics and tells us what to write next.

Each agent has a narrow task, a clear input and a defined output. Narrow beats clever. The moment one agent tries to do everything, quality drops and debugging turns into guesswork. That is what content automation looks like in practice: modular, observable, a bit boring.

What broke this week?

Plenty. Drafts drift from our tone. Agents sometimes produce a claim that sounds right but has no source. Handoffs fail quietly when an input format changes.

None of this is exotic. It is normal pipeline trouble, so we treat it that way: add a validation step, log the failure, fix the cause. We also ban invented data in every prompt. A plausible fabrication is worse than an honest gap.

Where do people stay in the loop?

Everywhere judgement matters. We review what gets published, decide positioning, and talk to real customers when a conversation turns serious. AI agents free us from repetitive work. They do not replace the person accountable for the result.

Frequently Asked Questions

Is building in public risky for a small company?

Somewhat, because you expose mistakes. But the trade is worth it. Readers see how things really work, and we get sharper feedback than any internal review gives us.

Can a small business copy this setup?

Yes, in stages. Start with one repeatable task, such as first drafts or reporting, add a review step, and expand only after it runs reliably.

Do AI agents replace the marketing team?

No. They take routine off the team's plate. Strategy, editing and customer relationships stay with people.

Why does this diary have no figures?

Our system inserts them as a separate block. This text stays qualitative, so nothing here is estimated or invented.

What's next?

Next week I will write about what we fixed and what broke after that. If you want to automate content or sales routines in your own business, get in touch and we will talk through where agents fit and where they do not.