This week in numbers (real, from our system)
- 🤖 AI agents running: 19
- 📝 Content published: 48 (blog RU 13, EN 19, Altezza 16)
- ⚙️ Generated programmatically: 4
- 📥 Leads in the system: 246 (+0 in the last 7 days)
Short answer: We run a content factory powered by AI agents that handle everything from keyword research and brief generation to drafting, SEO optimisation and publishing. Building in public means we share what works and what breaks — so small and mid-sized businesses can see a real, unfiltered picture of content automation before committing to it.
Look, most "AI content" posts are either breathless hype or vague hand-waving. Neither is useful if you are a business owner trying to decide whether to invest time and budget in automation. So we decided to do something different: build our own content engine at Arxitek, run it in the open, and write about it honestly every week — the wins, the failures, the weird edge cases.
This is that diary. Here is what we have built, how it actually works, and what you should think about if you want to do the same.
What Does "Building in Public" Actually Mean for a Business?
Building in public is a practice borrowed from the indie developer world. The idea is simple: instead of polishing everything behind closed doors and announcing a finished product, you share the process as it happens. The messy parts, the pivots, the things that did not work.
For a B2B company like Arxitek — focused on AI and business automation — this approach is not just a marketing tactic. It is a commitment to transparency with the exact audience we serve: owners of small and mid-sized businesses, marketing directors, and founders who are evaluating whether AI agents are worth the investment.
Here is the thing: our clients are smart. They can smell a polished case study that skips over the hard parts. When we build in public, we earn trust in a way that a glossy brochure never could. We show the scaffolding, not just the finished building.
It also forces discipline on us. If you know you are going to write about your process every week, you document it properly. You notice when something is not working. You fix it faster. Building in public is, in a strange way, a quality control mechanism as much as a content strategy.
For the businesses reading this: the same logic applies to you. If you are testing AI agents or content automation internally, consider narrating it — even just to your own team in a Slack channel or internal wiki. The act of explaining what you are doing surfaces assumptions you did not know you were making.
How Is Our Content Factory Actually Structured?
Let me walk you through the architecture without the jargon. Think of it as an assembly line where each station is handled by a specialised AI agent, and a human editor sits at the end of the line — not to rewrite everything, but to make the final call.
Stage One: Topic and Keyword Research
The first agent in the chain monitors search trends, competitor content, and our own analytics. Its job is to surface topics that are both relevant to our audience and have realistic ranking potential. It does not just pull a list of keywords — it clusters them by intent, separates informational queries from commercial ones, and flags seasonal patterns.
The output is a prioritised topic queue. A human — usually me — reviews it once a week and approves or rejects topics. This is not a rubber stamp. Sometimes the agent surfaces something that is technically high-traffic but wrong for our brand. Judgment still lives with the human.
Stage Two: Brief Generation
Once a topic is approved, a second agent builds the content brief. This includes the target keyword, secondary keywords, recommended structure, competitor angle analysis, and a list of questions the article must answer. The brief is the contract between the research layer and the writing layer.
This stage matters more than most people realise. A weak brief produces a weak article, no matter how capable the writing agent is. Garbage in, garbage out — that rule has not changed just because the writer is an AI system.
Stage Three: Drafting and SEO Structuring
The drafting agent takes the brief and produces a full Markdown draft — headings, body copy, meta description, FAQ section, internal link suggestions. It writes in the Arxitek voice: direct, first-person, practical, no corporate fluff.
This is where content automation pays its way. A draft that would take a human writer several hours to produce is ready in minutes. But — and this is important — the draft is a starting point, not a finished product. The agent does not know what happened in a client meeting last Tuesday. It does not have the institutional memory that makes a piece feel genuinely authored.
Stage Four: Human Edit and Fact-Check
A human editor reads every draft. They check for accuracy, add specific examples from our own experience, remove anything that sounds generic, and make sure the piece actually reflects how we think. This step typically takes less time than writing from scratch, but it is non-negotiable. We do not publish unreviewed AI output.
Stage Five: Publishing and Distribution
The final agent handles the mechanical work: formatting for the CMS, adding tags and metadata, scheduling the post, and queuing it for distribution across our channels. It also triggers a brief to the social media agent to create accompanying posts for LinkedIn and Telegram.
What Do AI Agents Handle Well — and Where Do They Fall Short?
Here is my honest assessment after running this system for a sustained period.
AI agents are excellent at:
- Consistency. They do not have bad days. The structure of every brief is the same quality whether it is the first of the week or the last.
- Scale. Once the workflow is set up, adding more topics does not require proportionally more human time.
- SEO mechanics. Keyword placement, meta structures, internal linking patterns — these are rules-based tasks that agents execute reliably.
- Research aggregation. Pulling together what is already published on a topic, identifying gaps, summarising competitor angles.
AI agents struggle with:
- Genuine opinion. An agent can simulate a point of view, but it cannot have one. The most valuable parts of our articles — the contrarian takes, the "here is what I actually think" moments — come from humans.
- Novelty. If something happened last week and is not in the training data, the agent does not know about it. Real-time awareness requires additional tooling.
- Brand nuance. Teaching an agent the difference between "our voice" and "a reasonable approximation of our voice" is an ongoing calibration process, not a one-time setup.
- Judgment calls. Should this topic be covered now or held for a better moment? Is this angle too controversial for our audience? These decisions stay with humans.
The honest summary: AI agents are brilliant amplifiers. They free the humans on the team from the mechanical, repeatable work so that human effort concentrates on the parts that actually require a human.
How Does Content Automation Affect SEO Results?
This is the question every marketing director asks, and I want to answer it without inventing numbers.
Content automation, done properly, improves SEO in several qualitative ways:
Consistency of publishing cadence. Search engines reward sites that publish regularly. When a human team is responsible for every step, publishing cadence is vulnerable to holidays, illness, competing priorities, and creative blocks. An automated pipeline removes most of those vulnerabilities. The calendar stays full.
Structural SEO discipline. Agents follow the brief. If the brief says every article needs a FAQ section, a meta description under a certain character count, and at least one comparison table — every article gets those things. Human writers, even good ones, sometimes skip the structural checklist when they are in a hurry.
Topic coverage breadth. One of the biggest SEO opportunities for small and mid-sized businesses is covering the full range of questions their audience asks — not just the obvious head terms, but the long-tail, conversational queries. Content automation makes it economically viable to cover that long tail systematically.
What automation does not fix: thin content, wrong audience targeting, or a site with technical SEO problems. Automation scales what you already have. If what you have is mediocre, you will get more mediocre content faster. The quality of the brief and the human edit layer are the real quality controls.
What Does the Weekly Diary Format Actually Look Like?
Each week we publish a short, honest update on how the content factory is performing. The format is deliberately simple:
- What ran this week: which content pieces went through the pipeline, any topics that were approved or rejected at the brief stage.
- What broke: any agent failures, prompt drift, formatting errors, or cases where the human editor had to do significant rework. We do not hide these.
- What we changed: any adjustments to prompts, workflow steps, or tooling. This is the engineering log.
- One observation: something we noticed that might be useful to anyone building a similar system.
For business owners considering content automation: this diary is the due diligence resource we wish had existed when we started. Read it before you build.
Is Content Automation Right for Your Business?
Let me be direct: content automation is not a fit for every business, and I would rather tell you that upfront than oversell it.
It is a strong fit if:
- You need to publish content regularly but do not have the headcount to do it manually at the required volume.
- Your content follows predictable formats — articles, product descriptions, SEO landing pages, email sequences.
- You have at least one person who can own the quality layer: reviewing drafts, maintaining prompts, making editorial calls.
- You are willing to invest time upfront in building the workflow correctly before expecting it to run smoothly.
It is a poor fit if:
- Your content is highly specialised, requires deep technical expertise, or depends on proprietary data that AI systems cannot access.
- You have no internal resource to manage and maintain the pipeline. Automation is not zero-maintenance.
- You want to "set it and forget it." That is not how this works. The pipeline requires ongoing calibration.
- Your brand voice is so specific and nuanced that the gap between agent output and finished article is larger than the time saved.
For most small and mid-sized businesses in the B2B space, the honest answer is: content automation handles the repeatable structure, and your people handle the judgment and expertise. That division of labour is where the real value lives.
Frequently Asked Questions
What is "building in public" and why does it matter for AI projects?
Building in public means sharing your development process openly — including failures and pivots — rather than only announcing finished results. For AI projects specifically, it matters because it builds trust with an audience that is rightly sceptical of polished AI marketing. It also creates a public record of what actually works, which is more useful to potential clients than any case study.
How many people does it take to run an AI content factory?
The honest answer is: it depends on the volume and quality bar you are targeting. The automation handles the mechanical work — research, briefing, drafting, formatting, scheduling. But you still need at least one person with editorial judgment to review output, maintain prompts, and make decisions about topics and angles. Automation reduces the headcount required, but it does not reduce it to zero.
Do AI agents replace human writers?
No — and that framing misses the point. AI agents handle the structural, repeatable parts of content production: keyword research, brief generation, first drafts, SEO formatting. Human writers and editors handle opinion, expertise, brand voice calibration, and quality control. The best content pipelines use agents to amplify human output, not to replace the humans. The pieces that perform best are always the ones where a human has genuinely engaged with the draft.
How do you maintain content quality when using content automation at scale?
Quality in an automated pipeline lives in three places: the brief (garbage in, garbage out), the human edit layer (non-negotiable), and the prompt maintenance routine (prompts drift and need regular recalibration). We also track which articles underperform and trace failures back to their source — was it a weak brief, a poor draft, or a missed edit? That feedback loop is what keeps quality from degrading over time.
What tools do AI agents typically use in a content automation workflow?
A typical stack includes a large language model for drafting and brief generation, a search or web-access tool for real-time research, a CMS integration for publishing, and an analytics connection for performance feedback. The specific tools matter less than the workflow design: how data flows between agents, where human checkpoints sit, and how failures are caught before they reach the published output.
What We Are Building Next
The content factory is one layer of a larger system. The next pieces we are working on — and will document here — are a sales assistant agent that handles initial inbound qualification, and an analytics agent that closes the loop between content performance and the topic queue. When those are running, the system will be genuinely self-improving: content that performs well informs the next round of topics, automatically.
We are also working on making the building-in-public format itself more structured — a proper changelog format, not just prose updates. The goal is that any business owner or marketing director reading this diary could adapt what we are doing to their own context without needing to reverse-engineer it from narrative.
Here is the thing: we are not doing this because it is easy. We are doing it because the alternative — closed-door development, polished announcements, no honest accounting of what broke — serves no one. Not our clients, not the wider community of people trying to figure out whether AI agents are worth the investment, and honestly not us either.
If you are building something similar, or if you are evaluating whether to start, get in touch. We would rather have a real conversation about your specific situation than send you a brochure. That is what building in public is actually about.