When Dustin McKissen decided to run his publishing company almost entirely with artificial intelligence, he wasn’t chasing a trend. He was solving a cash-flow problem. The result — a system he calls an “AI operating system” — has allowed his company, McKissen Media, to operate with near-zero human staff while producing content across multiple publications. It’s a case study that should make every mid-size media operator pay close attention.
McKissen, writing in TechRadar, laid out the architecture behind his approach in striking detail. He didn’t simply plug ChatGPT into a WordPress site. He built a layered system of AI agents — each assigned a specific role that would traditionally require a salaried employee — and connected them through a coordinated workflow that handles everything from content ideation to publication scheduling to performance analytics. The agents don’t freelance. They execute within tightly defined parameters, governed by what McKissen describes as detailed standard operating procedures translated into AI-readable instructions.
The concept isn’t entirely new. Businesses across sectors have been experimenting with AI agents — autonomous software programs that can complete multi-step tasks without constant human supervision. But McKissen’s implementation stands out because of its scope. He isn’t automating one part of the publishing pipeline. He’s automated nearly all of it.
The Architecture Behind the AI Newsroom
At the core of McKissen’s system is a series of specialized AI agents built on large language models, primarily from OpenAI and Anthropic. Each agent has a defined function. One handles research and topic generation. Another writes drafts. A third edits for tone, style, and factual consistency. Yet another manages SEO optimization. And a scheduling agent coordinates when and where pieces are published across McKissen Media’s portfolio of sites.
What makes this more than a gimmick is the connective tissue between agents. McKissen describes using automation platforms — tools like Make (formerly Integromat) and Zapier — to create handoff protocols between each AI agent. When the research agent identifies a trending topic with sufficient search volume and audience relevance, it passes a structured brief to the writing agent. The writing agent produces a draft that gets routed to the editing agent. And so on.
No single agent operates in isolation. That’s the key distinction between what McKissen built and the way most companies use generative AI today — as a tool an individual employee opens in a browser tab.
McKissen told TechRadar that the system required significant upfront investment in prompt engineering and workflow design. “The hardest part wasn’t getting the AI to write,” he explained. “It was getting the AI agents to talk to each other in a reliable, consistent way.” He compared the challenge to managing a team of new employees who are individually talented but need explicit instructions about how their work connects to everyone else’s.
The financial implications are hard to ignore. McKissen claims his operating costs dropped dramatically — he cited figures suggesting he runs his publishing operation for a fraction of what it would cost with a traditional editorial staff. For a small-to-mid-size publisher, that’s the difference between viability and shutdown.
But there’s a catch, and McKissen acknowledges it. Quality control remains a human responsibility. He personally reviews a percentage of the output, and he’s built in automated quality checks — essentially AI agents that audit other AI agents. Still, the final editorial judgment rests with him. The system doesn’t run unsupervised. Not yet.
This mirrors a broader pattern emerging across the media industry. Publishers are increasingly comfortable using AI for first drafts, headline testing, and distribution optimization. The Associated Press has used automated writing for earnings reports for years. BuzzFeed experimented with AI-generated quizzes. Forbes has explored AI-assisted content. But few have gone as far as McKissen in replacing the entire production chain.
Why This Matters Beyond One Small Publisher
The implications extend well past McKissen Media’s balance sheet. What he’s demonstrated — imperfectly, and at small scale — is a template. A playbook that any publisher with modest technical skills could replicate. The tools he used aren’t proprietary. They’re commercially available. OpenAI’s API. Anthropic’s Claude. Make. Zapier. Google Sheets for tracking. Nothing exotic.
That accessibility is precisely what makes this significant. The barrier to building an AI-powered publishing operation isn’t technology anymore. It’s knowledge — understanding how to decompose editorial workflows into discrete, automatable tasks, and then stitching them back together.
The media industry has been hemorrhaging jobs for two decades. Newsroom employment in the United States fell 26% between 2008 and 2020, according to Pew Research Center. The economic pressures haven’t eased. Digital advertising revenue continues to consolidate around Google and Meta. Local and regional publishers operate on razor-thin margins. Against that backdrop, a system that can produce publishable content at dramatically lower cost isn’t an intellectual curiosity. It’s an existential question for the industry.
Not everyone is enthusiastic. Critics argue that AI-generated content, no matter how well-orchestrated, lacks the judgment, sourcing rigor, and accountability that human journalists provide. And they’re not wrong — at least not entirely. AI agents can synthesize information and produce grammatically sound prose. They can’t knock on doors. They can’t cultivate sources. They can’t make the editorial judgment call about whether publishing a story will cause more harm than good.
McKissen’s publications, it should be noted, aren’t investigative journalism outlets. They’re content-driven media properties focused on business topics, career advice, and leadership — categories where AI’s limitations are less likely to produce serious errors than they would be in, say, political reporting or war correspondence.
That distinction matters enormously. The AI operating system McKissen built works in part because of what it’s asked to do. Scaling the same approach to hard news would be a fundamentally different — and far riskier — proposition.
Recent developments in the AI agent space suggest McKissen’s approach is about to get easier to replicate. OpenAI has been aggressively expanding its agent capabilities, and Anthropic recently introduced tool-use features for Claude that allow the model to interact with external software — browsers, code environments, file systems — with increasing autonomy. Google’s Gemini models are following a similar trajectory. The infrastructure for building multi-agent systems is maturing fast.
And the startup world has noticed. Companies like CrewAI, AutoGen (from Microsoft Research), and LangChain are building frameworks specifically designed to orchestrate multiple AI agents working in sequence or parallel. These tools abstract away much of the complexity that McKissen had to handle manually, which means the next publisher who tries this will have an easier time of it.
So where does this leave human journalists and editors? Probably not unemployed — but almost certainly redefined. The roles that survive in an AI-augmented publishing environment will be those that require judgment, relationships, and accountability. Editors who can evaluate AI output for accuracy and bias. Reporters who do original investigative work. Strategists who understand audience needs in ways that data alone can’t capture.
The commodity work — the daily content production that keeps publishing schedules full — is vulnerable. It has been for a while. McKissen just made the vulnerability concrete.
The Bigger Bet
What McKissen is really betting on is that the quality floor for AI-generated content will keep rising. That the agents will get better at research, better at nuance, better at mimicking editorial voice. And based on the trajectory of the underlying models, that’s not an unreasonable bet.
GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro — each generation of these models has shown measurable improvement in reasoning, factual accuracy, and stylistic range. If that curve continues, the gap between AI-generated and human-generated content in categories like business publishing will narrow further. Maybe to the point where most readers can’t tell the difference. Maybe to the point where it doesn’t matter.
McKissen isn’t claiming perfection. He’s claiming sufficiency — that the output is good enough, produced fast enough, and cheap enough to sustain a viable publishing business. For an industry that’s been struggling with all three of those variables simultaneously, that’s a proposition worth taking seriously.
Whether it’s the future of publishing or a cautionary tale depends entirely on what you think publishing is for. If it’s about filling pages and capturing search traffic, McKissen has cracked the code. If it’s about something more — accountability, investigation, the kind of reporting that holds power to account — then the AI operating system is a tool, not a replacement.
Probably both things are true at once. That’s what makes this uncomfortable.
WebProNews is an iEntry Publication