This week in numbers (real, from our system)
- 🤖 AI agents running: 19
- 📝 Content published: 45 (blog RU 13, EN 17, Altezza 15)
- ⚙️ Generated programmatically: 4
- 📥 Leads in the system: 236 (+0 in the last 7 days)
Figures as of 2026-07-29 — computed by code from the DB and files, no manual entry.
Short answer: At Arxitek we run a content factory powered by AI agents that handle research, drafting, SEO optimisation, internal linking, and analytics reporting — with a small human team setting direction and approving output. This diary documents what actually works, what breaks, and how the system evolves week by week, so other business owners can learn from it without the trial-and-error cost.
Let's be honest: most "AI content" posts are either breathless hype or vague theory. This one is neither. This is a weekly engineering diary — raw, opinionated, and grounded in what we actually built and run at Arxitek. We automate content production using AI agents, and we're doing it in public so you can see the real mechanics, not a polished case study.
If you run a small or mid-sized business and you're wondering whether content automation is worth the investment, or if you're a marketing director who needs to explain to a sceptical CEO why AI agents are not just a toy — read on. This is the honest version.
What Does "Building in Public" Actually Mean for Us?
Building in public is a commitment to radical transparency. Instead of waiting until everything is perfect and then publishing a triumphant post-mortem, we share the process as it happens — the decisions, the failures, the pivots, and the small wins. For a technology company like Arxitek, which builds AI automation systems for businesses, this is both a marketing strategy and an accountability mechanism.
Here's the thing: when you build in public, you cannot hide behind polished metrics. You have to describe what you actually did this week, what the system got wrong, and what you changed. That discipline is uncomfortable, but it forces better engineering. It also builds trust with exactly the audience we care about — business owners and marketing directors who are tired of vendor promises and want to see real systems in action.
For us, "building in public" means publishing this diary every week. It means showing the architecture of our content automation pipeline, explaining why we chose certain tools and rejected others, and being upfront when an AI agent produces garbage output that needs a human to fix it. It means treating our own blog as a living laboratory, not a brochure.
The secondary benefit is SEO. A weekly diary with consistent publishing cadence, rich internal linking, and genuine technical depth is exactly what search engines and AI answer engines reward. Every post in this series cross-links to the others, building topical authority around AI agents and content automation. That is not an accident — it is part of the architecture.
So when I say we are building in public, I mean we are doing the work, documenting the work, and letting the documentation do its own marketing. No fabricated testimonials, no invented metrics, no "a client saved X hours" stories. Just the engineering.
How Are Our AI Agents Structured Inside the Content Factory?
The content factory is a multi-agent pipeline. Each agent has a narrow, well-defined job. That is the first principle we learned: generalist agents that try to do everything produce mediocre output at every stage. Specialist agents that do one thing well, chained together with clear handoffs, produce consistently usable output.
Here is how the pipeline is structured at a high level:
Research Agent
The research agent monitors industry sources, competitor content, search trends, and internal analytics. It does not write anything. Its only job is to surface topics worth covering, identify gaps in our existing content, and produce a structured brief — target keyword, search intent, competitive landscape, key questions the article must answer. The brief goes into a shared queue.
This is the stage where human judgment matters most. A marketing director reviews the queue, prioritises topics, and approves briefs before they move forward. The agent saves hours of manual research; the human ensures the editorial direction stays coherent.
Drafting Agent
The drafting agent takes an approved brief and produces a structured first draft. It follows a strict template: short-answer block for AI snippet capture, intro, H2 sections with minimum depth requirements, FAQ, and conclusion. The template is enforced at the prompt level and validated programmatically before the draft is accepted.
The draft is not published automatically. It goes to a human editor — in our case, a content lead — who checks for accuracy, voice consistency, and any claims that need verification. The agent handles the structural and informational scaffolding; the human handles nuance and brand voice.
SEO and Internal Linking Agent
Once a draft is approved, the SEO agent runs through it. It checks keyword density, heading structure, meta description length, and — critically — internal linking opportunities. It queries our content index, finds relevant existing articles, and suggests anchor text and placement. A human confirms the links before they go live.
This agent alone saves a meaningful amount of time per article. Manual internal linking is one of those tasks that is easy to skip when you are busy, which is exactly why content silos form on most business blogs. Automating the suggestion layer removes the friction.
Analytics and Reporting Agent
The analytics agent pulls performance data weekly — rankings, impressions, clicks, time on page, and conversion events. It generates a structured report that highlights what is improving, what is stagnating, and what needs attention. It flags articles that have dropped in ranking and recommends whether to update, consolidate, or redirect.
This is the feedback loop that makes the whole system self-improving. Without it, you are publishing into the void and hoping for the best. With it, you have a clear editorial backlog of content that needs work, prioritised by potential impact.
What Does a Real Week in the Content Factory Look Like?
Let me walk you through a typical week, without invented numbers, because the honest qualitative picture is more useful than fabricated metrics.
Monday: The research agent delivers its weekly brief queue. I spend roughly half an hour reviewing it — skimming the briefs, rejecting topics that do not fit our editorial direction, flagging two or three for immediate drafting, and moving the rest to a backlog. This is the most important half-hour of the content week. The agent does the legwork; I make the calls.
Tuesday–Wednesday: Approved briefs go to the drafting agent. Drafts come back within minutes, not hours. The content lead reviews each one, edits for voice and accuracy, and either approves or sends back with comments. The agent does not get defensive about edits — it just re-drafts. This back-and-forth is faster than working with a junior freelancer and produces more consistent structure.
Thursday: The SEO agent runs on approved drafts. Internal linking suggestions are reviewed and confirmed. Meta descriptions are checked. Articles are scheduled for publication.
Friday: The analytics agent delivers its weekly report. We review what moved, what did not, and update the editorial backlog accordingly. Any articles flagged for update go into next week's queue.
The whole editorial operation runs on a small team. The agents handle the volume and the repetitive cognitive work. The humans handle judgment, voice, and strategy. That is the division of labour that actually works.
What Breaks, and How We Fix It
Building in public means talking about failure. Here is what breaks regularly in our content factory and what we do about it.
Hallucinated facts. AI agents will occasionally state things with confidence that are simply wrong — a statistic that does not exist, a tool feature that was deprecated, a claim that sounds plausible but is not accurate. Our mitigation is a strict editorial rule: any factual claim that is not common knowledge must be verified by the human editor before publication. The agent is not a source of truth; it is a drafting tool.
Voice drift. Over time, without careful prompt maintenance, agents start drifting toward a generic corporate tone. We fix this by regularly updating the voice guidelines in our prompt library and running spot-checks on recent output. When we notice drift, we treat it as a prompt engineering problem, not an agent failure.
Over-optimised structure. Sometimes the SEO agent recommends adding keywords in ways that make the text feel mechanical. The human editor has override authority on every SEO suggestion. The rule is: if it reads awkwardly, it does not go in, regardless of what the agent recommends.
Stale briefs. The research agent works from data that is, at best, a few days old. Fast-moving topics can become stale between brief generation and publication. The content lead is responsible for checking whether a brief's topic is still relevant before approving it for drafting.
None of these are catastrophic failures. They are the normal friction of running a complex system. The discipline of building in public means we document them rather than pretend they do not happen.
How Does This Apply to Small and Mid-Sized Businesses?
Here is the honest answer: you do not need to build what we built from scratch. The architecture we use — research, drafting, SEO, analytics, all agent-driven with human checkpoints — is replicable with off-the-shelf tools and a clear workflow design.
For a small business owner or a marketing director managing a lean team, the most important insight from our experience is this: the bottleneck in content marketing is almost never creativity. It is consistency and process. Most businesses publish sporadically because the process is too manual and too dependent on individual energy. AI agents solve exactly that problem — not by replacing the creative judgment, but by handling the repetitive scaffolding that makes consistency possible.
A realistic starting point for a small business might look like this:
- A single drafting agent configured with your brand voice and editorial template
- A simple brief approval process that takes fifteen minutes a week
- A human editor who reviews and approves before publication
- A monthly analytics review to close the feedback loop
You do not need a multi-agent pipeline on day one. You need a process that is more consistent than what you have now. The agents come in to make that process scalable.
| Stage | Without AI Agents | With AI Agents |
|---|---|---|
| Topic research | Several hours, manual | Agent delivers structured brief queue |
| First draft | Hours per article | Minutes per article, human edits |
| SEO & internal linking | Often skipped | Agent suggests, human confirms |
| Analytics review | Monthly at best | Weekly automated report |
| Publishing cadence | Sporadic | Consistent, process-driven |
The table above is not a promise of specific time savings — actual results depend on your team, your tools, and your content complexity. It is a structural comparison of what changes when you introduce content automation into a real editorial workflow.
Frequently Asked Questions
What is "building in public" and why does it matter for AI projects?
Building in public means sharing your process, decisions, and failures openly as you work — rather than only publishing polished results after the fact. For AI projects, it matters because it creates accountability, builds audience trust, and generates genuinely useful content that reflects real engineering experience rather than theoretical advice. It is also a strong long-term SEO strategy because it produces consistent, original, experience-based content.
How do AI agents handle content automation without losing brand voice?
AI agents handle content automation by following detailed prompt instructions that encode your brand voice, editorial rules, and structural templates. The key is treating voice guidelines as a maintained asset — not something you write once and forget. Regular spot-checks, prompt updates when drift is detected, and mandatory human editorial review at every publication stage are what keep the output on-brand. Automation handles volume; humans protect voice.
Is content automation suitable for small businesses, or only for large teams?
Content automation is arguably more valuable for small businesses than for large ones, precisely because small teams have less capacity for repetitive manual work. A small business with one or two people responsible for marketing can use AI agents to produce consistent, well-structured content at a cadence that would otherwise require a much larger team. The key is starting simple — one agent, one clear workflow, one human checkpoint — and building complexity only as needed.
What are the biggest risks of using AI agents for content production?
The main risks are factual inaccuracy (agents can state wrong things confidently), voice drift over time, over-optimised or mechanical-sounding text, and the temptation to remove human review in the name of speed. All of these are manageable with clear editorial rules, regular prompt maintenance, and a firm policy that no AI-generated content is published without human approval. The risk is not the technology — it is the process around it.
How long does it take to set up an AI-driven content factory?
A basic setup — a drafting agent with a brand voice template, a brief approval step, and a human editorial review — can be operational within a few weeks for a small business. A more sophisticated multi-agent pipeline with research, SEO, and analytics agents, integrated into your CMS and analytics stack, typically takes longer and benefits from working with a specialist. The investment in setup pays back through consistency and reduced manual effort over time.
Conclusion: Why We Will Keep Doing This
Building in public is uncomfortable. It means admitting when the research agent surfaces a terrible brief, or when a draft comes back so generic it needs to be almost entirely rewritten. It means publishing a diary entry on a week when the system produced nothing remarkable.
But here is the thing: that discomfort is the point. The discipline of documenting the real process — not the idealised version — is what makes this useful to you. If you are a business owner or a marketing director trying to figure out whether content automation is worth building, you deserve an honest account, not a sales deck.
We will keep publishing this diary every week. The system will keep improving. And you will be able to follow along, take what is useful, and skip what is not.
If you want to talk through how a content automation setup might work for your business specifically — the architecture, the tools, the human workflow around it — get in touch with the Arxitek team. We are happy to have a practical conversation, no pitch required.
Top comments (0)