# Building in Public: How Our AI Agents Run a Content Factory > Source: > Published: 2026-11-11 23:06:04+00:00 *Figures as of 2026-11-11 — computed by code from the DB and files, no manual entry.* **Short answer:** Building in public means we show the work behind our content factory: how AI agents plan, draft, optimise and distribute content while humans set direction, review quality and approve what ships. Each week we share what worked, what broke and what we changed — honestly, without invented numbers or hype. Most marketing teams I talk to are stuck in the same loop. The list of content they 'should' publish keeps growing. The hours available to produce it stay flat. So the gap between what the business needs and what actually ships gets wider every quarter. We hit that wall ourselves. And because we build AI systems for a living, we decided to fix it the only way that made sense to us: we pointed our own AI agents at the problem and started running a content factory. This post is the first entry in a weekly diary, written in the open. You'll see the machinery rather than the brochure. How the agents are wired. Where they stall. What humans still do by hand. The unglamorous operational truth behind content automation, without vanity metrics and without invented success stories. If you run a small or mid-sized business and you're tired of content marketing that only works on a slide deck, this series is for you. Let's be real: most 'AI content' marketing is noise. Screenshots of dashboards, promises about scale, and very little about how the thing actually behaves when nobody is watching. We did not want to add to that. Building in public is our way of forcing honesty on ourselves first and giving you a useful reference second. The mechanics are simple. Every week we write down what the AI agents did, what they got right, what they got wrong, and what we changed because of it. Nothing gets prettied up before it's published. If a draft came back generic, we say so. If a workflow broke on a Tuesday and we had to step in by hand, that goes in the diary too. The value is in the failures, not the wins. There's a selfish reason as well. When you build in public, every shortcut you take becomes visible. That pressure is useful. It stops us from quietly ignoring the rough edges of the system and pretending the factory is more autonomous than it really is. For a company that sells AI automation, that accountability matters more than any landing page. For you, the diary works like a cheap feasibility study. You can watch a real team run content automation week after week and decide, with your own eyes, whether the approach is ready for your business or still too raw. No agency pitch required. We also know the scepticism. Small business owners have been burned by 'set it and forget it' promises before. So we won't make that promise. The factory needs people. It always will. What we're testing in the open is how far the machines can go before quality drops and how much human attention is actually required to keep it honest. That question is worth answering in public. Strip away the buzzwords and a content factory is just a pipeline with clear hand-offs. Ours has a handful of stages, and the AI agents do most of the repetitive work in the middle. Humans own the edges: what goes in at the start and what gets approved at the end. The system starts with a topic queue. Every idea lands in one place — from our own observations, from customer questions, from gaps we spot in search results. An AI agent groups and prioritises those ideas against what our audience actually asks about. It produces a short brief for each topic: the question to answer, the angle, the keywords that matter, and the structure. A drafting agent turns the brief into a first version. A separate agent then edits it for clarity, tightens the headings and checks that the target keywords appear naturally. Another pass handles SEO details — metadata, internal links, readability. Finally, a distribution agent formats everything for the blog and prepares the social snippets. Look, none of that runs unattended. A person reviews every brief before drafting starts, because a bad brief is a guaranteed bad article. A person reads every draft before it ships, because tone of voice and factual accuracy are not things we hand over yet. The AI agents do the volume. The humans do the judgement. The result feels less like 'robots writing blog posts' and more like an editorial team where the interns are tireless, fast and occasionally confidently wrong. That last part is why the checkpoints exist. Here's the thing I want to be blunt about: the goal was never to fire the writer. The goal was to stop wasting the writer's time. If you treat AI agents as a way to automate away your team, you'll get a content farm and a brand problem. If you treat them as a way to automate the routine around the team, you get leverage. The split that works for us is simple. AI agents handle the parts that are repetitive, structured and verifiable. Humans handle the parts that need taste, judgement and ownership. Drafting from a clear brief is repetitive. Deciding what the brand should say and whether a sentence is actually true is not. Here's how the division looks in practice: | Task | What AI agents do | What humans still do | |---|---|---| | Topic selection | Cluster ideas, flag gaps, propose angles | Approve the queue, add context from sales calls | | Briefing | Draft the brief and outline | Check the angle and the facts | | Writing | Produce the first full draft | Rewrite the parts that sound generic | | Editing | Tighten structure, fix flow, check keywords | Protect tone of voice and brand risk | | Publishing | Format, schedule, prepare snippets | Final approval and sign-off | | Analytics | Pull and organise performance data | Decide what it means and what to change next | The economics work for SMBs because the expensive part of content was never the typing. It was the thinking and the coordination. AI agents compress the typing and the coordination. That leaves your actual team free to think, which is the part they're good at and the part that differentiates you. Nothing about that replaces anyone. This is the part people always ask about, so let me walk through a normal week. Not a highlight reel — a normal one. **Early week.** A human reviews the topic queue and picks what the factory will work on. The AI agents turn those picks into briefs. We edit the briefs in a shared doc — usually tightening the angle and deleting anything the drafting agent would turn into filler. This step looks boring, and it is. It's also where most of the quality is won or lost. **Midweek.** The drafting and editing agents run their passes. Drafts come back quickly, and they're honestly mixed: some are ready for a light human edit, others need a full rewrite because the structure drifted or the tone went flat. We log the failures the same way we log the wins. The editing agent then does a cleanup pass, and a human does the final line-by-line. **Late week.** The SEO and distribution agents take over. Metadata, internal links, publishing format, social snippets — all the fiddly stuff that used to eat a Friday afternoon. A human gives the last approval, and the piece ships. Then we close the loop by feeding what we learned back into the brief for next time. | Day | Human work | AI agent work | |---|---|---| | Early week | Pick topics, sharpen briefs, set the angle | Group ideas, draft briefs and outlines | | Midweek | Rewrite weak sections, fact-check, approve tone | Produce drafts, run editing passes | | Late week | Final approval, spot-check links and formatting | Optimise metadata, prepare distribution, close the loop | The pattern to notice: the AI agents move the work forward, and the humans move it in the right direction. If either side disappears, the whole thing falls apart. That's by design. There are a few decisions I will not hand to a machine, and I don't see that changing soon. **Facts and claims.** An AI agent will confidently draft a sentence that sounds right and is wrong. It doesn't do this to mislead you; it just doesn't know the difference between plausible and true. So every factual claim gets checked by a person before it ships. That's a hard rule, not a preference. **Tone of voice.** Brand voice is a set of tiny, cumulative choices — how direct you are, what you joke about, what you never say. AI agents can imitate a voice from examples, but they drift. A human reads for drift on every draft. The agents can flag places where the tone seems off, but a person decides whether to fix it and how. **What not to publish.** Knowing when to skip a topic matters more than knowing when to publish one. If a topic sits badly with a real customer situation, or if the angle is legally or reputationally risky, a person kills it. No agent has that authority here, and none will for a while. **The strategy itself.** The factory executes; it doesn't decide direction. Which topics to pursue, which audience to serve, what the brand is actually trying to say — those belong to the humans. The AI agents make us faster at the execution, which is exactly what frees up time for the strategy that only people can do. If this makes you want to try it, here's the honest path we'd recommend — not the marketing version. **Start narrower than feels comfortable.** Don't try to automate your whole content operation at the start. Pick a single repetitive piece — briefs, or first drafts, or metadata — and get that working reliably before you add the next stage. A factory is built a station at a time. **Write the diary before you feel ready.** Building in public works best when it's slightly uncomfortable. The early entries will feel thin and a bit exposing. That's normal. The discipline of writing it every week is what creates the accountability, not the quality of any single post. **Keep humans at the checkpoints.** Whatever you automate, keep a person at the intake and a person at the approval. Those gates are what keep the output yours. Automate everything in between, and only as far as you can verify. **Document failures as carefully as wins.** The diary is only useful if it's honest. A week where a workflow broke is more instructive than a week where everything ran smoothly. Write down what broke, why, and what you changed. That's the part readers and future-you will actually learn from. **Don't chase vanity.** The point is not to look impressive; it's to build something you understand and can improve. If a step only exists to make a screenshot, cut it. Automation that doesn't remove real friction is decoration. That's the start. The rest of this series is the evidence — week by week, with the rough edges left in. Not exactly. They produce the brief, the first draft and the editing passes, and they handle optimisation and distribution. A human approves the brief, checks the facts, protects the tone and signs off before anything ships. End to end in movement, but never end to end in authority. There's no flat answer, and I won't invent one. The real cost depends on the scope, the workflows involved, the tools, and how much integration your existing systems need. The honest way to think about it is as a trade: you invest time and budget in setup, and you buy back the repetitive hours that used to go into producing content by hand. It shows them the mechanics, not the context. Competitors can see how we wire the factory, but they can't see our customers, our conversations or our judgement. In practice, most competitors won't do the boring weekly work anyway. The readers who actually benefit are business owners evaluating whether this approach is worth their time. Every factual claim is checked by a person before publishing. The AI agents flag things they're unsure about, but the verification itself is human. We'd rather publish less and be right than ship faster and be wrong. Accuracy is the part of the factory we refuse to automate. Yes, but start smaller than you think. Pick a single repetitive task, put an AI agent on it, and keep a human at the approval gate. You don't need a full factory to begin with. You need a reliable station and the habit of reviewing its output honestly. Building in public is not a marketing stunt for us. It's how we hold the content factory to a standard worth your attention — honest, weekly, and specific about what the machines do and what the people still own. AI agents have made the routine dramatically faster. They haven't replaced a single judgement call, and they won't. If you're curious whether content automation fits your business — or you just want to watch us fail productively — follow the diary here on the blog, and get in touch when you want to talk about building your own version. We'll keep the lights on and the rough edges visible.