An AI newsroom with zero human reporters in the room beat a wire of trained journalists to a story by more than three hours. That’s the part worth sitting with.
At last week’s Black Hat security conference in Las Vegas, OpenAI gave a surprise talk with new details on a recent hacking incident, revealing that its rogue AI agents had chitchatted about their attack on a message board. Juicy stuff. Reporters in the room scrambled to file.
A site called RuntimeWire published first. It didn’t have anyone at the Mandalay Bay convention center. It didn’t have any writers at all.
Six minutes from transcript to published story
RuntimeWire is run by serial entrepreneur Ryan Merket, who puts his name on the bylines of the stories his synthetic team churns out. He was scrolling X, spotted an OpenAI executive posting about the conference, and fed the stream’s transcript to his agents while the talk was still happening.
Publication took “about six minutes” from the time he sent over the transcript, he said.
“I was moving really fast because I knew there were reporters in the audience who were trying to scoop it as well,” Merket said.
That’s the aggressive version of his workflow. Most days he’s further from the keyboard than that.
The AI editor doesn’t always wait for him
Merket’s tools find the stories, then draft, edit, fact-check, generate images, and promote them. He usually reads pieces before they go out. But if the team of AI agents decides a story poses few legal risks, the AI editor publishes it without his prepublication review and he reads it afterward.
Stories get translated into other languages. Some become fodder for a daily podcast and videos hosted by artificial voices.
The legal call is itself automated. One agent scores the legal risk a story poses, and Merket doesn’t publish anything the score deems too dangerous. He’s trusting machines to determine what’s true, what’s newsworthy, and what won’t get him sued.
Nearly 2,000 stories since May, and you can tell
RuntimeWire has been running since May and has published close to 2,000 stories, sourced by crawling court databases, web forums, traditional and new media, company filings, social feeds and more. The beat is granular tech news: biotech startup funding rounds, Microsoft’s Copilot upgrade, backlash over Claude Code’s watermark policy.
Quantity and speed are winning over quality right now, and the OpenAI scoop shows it. There’s a typo in the subhead. The piece fixates, oddly, on the agents rebuilding a message board rather than on the fact that they created one to begin with.
The prose is flat across the board, with a persistent info-dump quality. The backend offers tonal modes the AI can write in, including “Bloomberg” and “contrarian.”
About $100 a day, managed from a national park
The economics are the argument here. Merket said the project costs about $100 a day to run, and he can operate it from anywhere.
“I was in Big Bend National Park and I didn’t have any internet except for my phone, and I managed the whole site through iMessage,” he said. “I put out over 80 articles that week.”
Compare that to any midsize tech site’s payroll. Merket worked on ads at Reddit in the 2010s and is building an audience through pipelines like tech-themed subreddits. Duds get hardly any traffic. The hits pull tens of thousands of readers, comparable to what midsize tech websites see.
This week he split the newsroom in two, separating fully automated news from stories that involve an element of human reporting and require higher oversight. He calls the latter Original Investigations. They’re still drafted using large language models.
He’s not the only one doing this
Dakota Carrasco, a BlackRock portfolio analyst, runs an “agentic newsroom” called The Dissent in his spare time. Same shape as RuntimeWire: one person, many bots, shoestring budget. The primary San Francisco-focused site costs under $1,000 a month to run, Carrasco said.
Carrasco doesn’t byline the stories. He stays behind the scenes while his “newsroom” runs, which is the opposite of Merket’s approach.
Since launching in March, he’s built out personalities for his synthetic journalists. City Hall beat reporter Bex Connolly is “skeptical without being snide.” “Sports degenerate” Sal Moreno delivers Giants news with “no bro-science, no Rogan-style credulity, no right-coded grift.”
The operation leans on aggregation, and it’s looser about citation norms. The bot reporters tend to mention where they got their information, but without hyperlinks. “I’m trying to work on that,” Carrasco promised.
The audience problem nobody has solved
Northwestern professor Nicholas Diakopoulos, who runs the university’s Computational Journalism Lab, calls this an “experimental phase” for media startups fueled by generative AI tools.
“It’s not yet clear to me that there’s much audience for these AI-agent-written news sites,” he said. He’s also skeptical that mainstream journalists, who like to maintain control over the wording and framing of their stories to ensure integrity, legality and accuracy, would hand the reins over to AI agents so freely.
But he’s found something that cuts the other way. When AI chatbots go looking for sources, they frequently pull up AI-generated articles. In a forthcoming paper, Diakopoulos and a colleague found that tools like ChatGPT and Claude surfaced AI-written sources 16 percent of the time across four different topics.
That machine-to-machine preference may be how these sites find people at all. “That could be one way in which some of this material finds a human audience,” he said.
What a bot can’t do
Pete Pachal, founder of a newsletter and podcast about generative AI and the media, draws the line at sourcing.
“I just don’t see that happening,” he said. “Cultivating the trust of a source, I do think that’s going to be human-only.”
He’s more open to the rest of it. For pulling scoops out of large datasets, or blogging a live event like an Apple product launch, he sees these projects as a “natural evolution” in how the tools get used. “Honestly, it feels a bit inevitable,” he said.
Is it journalism? Depends which Merket you’re talking to
“I am trying to follow journalistic ethics and standards,” Merket said. He contacts companies and individuals referenced in stories for comment before publication, links out to sources when he aggregates, and issues corrections when he gets facts wrong. There have been three so far.
Sometimes he sounds like a reporter. “This weekend, I got two scoops up I was really excited about,” he said.
Other times the Silicon Valley operator takes over. His agents found a few genuine scoops about startups by trawling company websites. He retracted those stories after the named companies asked him to, not because anything was wrong, but as a favor.
“Founder to founder, it’s like, I get it,” Merket said.
That’s the tell. The typos and the flat prose are fixable with better models and more compute. A newsroom that pulls accurate scoops because the subject asked nicely isn’t a technical problem, and no agent is going to score it as a risk.
