I Built a Publication That Runs Itself. Here's What That Actually Means.
- Rich Washburn
- 2 days ago
- 5 min read
On autonomous content flywheels, AI editorial staff, and what happens when the bottleneck isn't the work — it's the decision.
Every morning at 7:00 AM, before I've had coffee, a report lands in my inbox. It's not a newsletter someone else wrote. It's not a Google Alert digest. It's a structured intelligence briefing — researched, scored, and written by an AI that's been scanning the internet since before I woke up. It tells me what moved in the space I cover, ranks each story by signal quality versus hype, and presents me with a short list of the most important things happening that day. My job, at that point, is to read it and reply. That's it.
I might say: "Run articles on items two and four. Skip the rest."
I might add context: "For item two, angle it toward enterprise security teams, not researchers."
Or I might just reply with a YouTube link and say: "Embed this in item four."
Within minutes, fully researched, original articles are written, cover images are generated, and everything is live on the site — including that YouTube video playing inline, not linked away, actually embedded and playable right there in the article body. I didn't write a word. I made two editorial decisions. That's the product.
What's Actually Running
I want to be precise about what this is and what it isn't, because a lot of people hear "AI content" and picture low-quality SEO slop with no editorial judgment. That's not this.
The system is built around a few distinct capabilities that work together as a flywheel.
Research first. Every morning the system scans primary sources, institutional announcements, research publications, and tech press across the specific domain it covers. It doesn't just find headlines — it cross-references multiple sources to understand what's actually being said versus what's being amplified. It scores each story on two dimensions: signal quality (how substantive is the underlying development?) and hype level (how much attention is it getting relative to what it deserves?). That scoring shows up as metadata on every published article, so readers can orient themselves quickly.
Then the briefing. The research gets packaged into a structured daily report that arrives in my inbox. I read it, I weigh in however I want, and I reply. The reply is the editorial layer. Sometimes I have opinions. Sometimes I just say go. Either way, that human checkpoint is what keeps the output from being a fire hose of automated content nobody asked for.
Then execution. Once I give direction, the system handles everything: research depth, article structure, writing, image generation, publishing. If I drop in a link to a relevant video or a source I want included, it gets embedded — not linked, embedded. The article renders it inline. That's the difference between a platform and a pipeline.
Then the site. The publication itself is a full editorial product — not a template, not a CMS skin. It has a category taxonomy, author bylines, featured article logic, a glossary, a company tracker, a newsletter capture system, an ad management layer, and an embedded AI assistant that can answer reader questions in real time. Analytics and search data feed back into the system automatically.
The whole thing runs continuously. I'm the editor. The machine is the staff.
The Team I Didn't Have to Hire
Think about what it actually takes to run a professional publication.
You need someone to monitor the space and identify what's worth covering. You need a writer who understands the subject matter well enough to produce something a technical audience will respect. You need a designer for cover images and visual formatting. You need an SEO person managing metadata, tags, categories, and keyword targeting. You need someone handling social distribution. You need analytics reviewed and acted on. You need someone keeping the site itself configured and running.
That's seven people. In a traditional media company, that's a team, a budget, an office, benefits, and a six-month runway before the first article goes live. Every one of those functions is running on the platform I built.
Not as a replacement for human creativity — as infrastructure for it. The research is faster than any human analyst. The publishing is instant. The image generation is immediate. The editorial judgment is still mine.
I'm not working less. I'm working at a different layer.
What I'm Actually Selling
Here's where I'll be deliberately incomplete, because we're building this into a product and I'm not going to hand-draw the blueprint in public.
But the concept is this: a full-stack content intelligence operation — the site, the AI editorial staff, the automation, the integrations — packaged and operated for a specific vertical. You tell us the space you want to cover. We build the publication. The AI runs it. You make editorial calls.
For some clients that looks like a daily reply to a briefing email. For others it's a Telegram message: "Write about the IBM announcement from this morning." The system handles the rest.
What you end up with is a publication that looks, reads, and performs like it has a full team behind it — because functionally, it does. The team just doesn't clock in.
The bottleneck in content, in most organizations, has never been lack of ideas. It's execution capacity. You know what you should be publishing. You don't have the bandwidth to produce it, the budget to staff it, or the infrastructure to distribute it consistently. That's the gap this closes.
The Deeper Point
I've been writing lately about synthetic labor — about assembling capability around a problem rather than headcount. About how the bottleneck in knowledge work has shifted from execution to clarity of intent. A publication that runs itself is the most direct demonstration of that I've built.
The machine does the research because research is a pattern-matching problem at scale, and machines are better at it than humans. The machine does the writing because, with the right structure and the right domain knowledge, it can produce something genuinely useful to a specific audience. The machine does the publishing because publishing is logistics, and logistics should be automated. I do the editorial direction because that's judgment, and judgment is still the job.
That division of labor — machine handles execution, human handles intent — is what makes this different from automation in the old sense of the word. This isn't a robot doing a repetitive task faster. It's a collaborator that handles the parts of the work where speed and consistency matter, so the person driving it can operate at the level where judgment actually makes a difference. The publication exists. It's live. It's growing.
Rich Washburn is a technologist and strategist working at the intersection of AI, infrastructure, and capital. He is Managing Partner and Chief AI Officer at Eliakim Capital and CIO of Data Power Supply.
Comments