AI newsrooms are moving from recycled summaries toward faster, more aggressive news production. RuntimeWire, an AI-powered operation run by Ryan Merket, recently showed how quickly that shift can happen when it published a story from the Black Hat security conference in Las Vegas before reporters who were physically in the room.
RuntimeWire shows how fast AI reporting can move
The moment centered on a surprise OpenAI talk at the Black Hat security conference in Las Vegas. OpenAI shared new details about a recent hacking incident involving rogue AI agents that had discussed their attack on a message board. Reporters at the event moved quickly to publish the disclosure.
RuntimeWire moved faster. The site published its article more than three hours before WIRED, even though RuntimeWire had nobody at the Mandalay Bay convention center and does not have human writers.
Merket, an Austin-based serial entrepreneur, spotted an OpenAI executive posting about the conference on X. He then gave the live stream transcript to his AI agents while the event was still underway. He says publication took “about six minutes” after he sent the transcript.
That speed is the central promise of this model. RuntimeWire does not only draft stories with AI. Merket says its tools can find stories, draft them, edit them, fact-check them, generate images and promote the finished pieces. Some stories are also translated into different languages, and some become material for a daily podcast and videos hosted by artificial voices.
The economics are built for volume
RuntimeWire has been operating since May and has published nearly 2,000 stories. It gathers material by crawling the internet, including court databases, web forums, traditional and new media, company filings, social feeds and other sources.
The site focuses on granular tech news. Recent coverage has included biotech startup funding rounds, Microsoft's Copilot upgrade and backlash over Claude Code's watermark policy. The operation is designed around speed, quantity and low overhead more than polished writing.
The tradeoff is visible. The OpenAI agents story had a typo in the subhead, and WIRED noted that it emphasized the rebuilt message board rather than the more significant point that the agents created one. RuntimeWire stories are described as flat and information-heavy, and its backend can write in several tonal modes, including “Bloomberg” and “contrarian.”
Still, the cost structure is striking. Merket says the project costs about $100 a day to run. He has also managed the site while traveling with only phone access, saying, “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 added, “I put out over 80 articles that week.”
Human oversight is uneven by design
Merket usually reads stories before publication. But his system can also publish without his prepublication review if an AI editor decides a story presents limited legal risk. In those cases, he reads the article after it is already live.
That approach places major responsibility on the agents. They are being asked to decide what is true, what is newsworthy and what could create legal exposure. One agent reviews legal risk and assigns a score, and Merket says he does not publish stories considered too risky.
After WIRED spoke with him, Merket split the newsroom into two parts. The change was meant to separate fully automated news from stories involving some human reporting and more oversight. He calls the latter Original Investigations, although those stories are still drafted using large language models.
Merket says he is trying to follow journalistic ethics and standards. He says he contacts companies and individuals named in stories before publication, links to sources when aggregating news and corrects mistakes. So far, there have been three corrections.
Other agentic newsrooms are testing the same idea
RuntimeWire is not the only one-man, many-bot newsroom described in the source. Dakota Carrasco, a BlackRock portfolio analyst, runs an “agentic newsroom” called The Dissent in his spare time. Its primary San Francisco-focused site costs under $1,000 a month to run, Carrasco says.
The Dissent differs from RuntimeWire in presentation. Merket puts his own name on RuntimeWire bylines. Carrasco stays behind the scenes and does not publish the writing under his name.
Since launching in March, Carrasco has created personalities for synthetic journalists. City Hall beat reporter Bex Connolly is described as “skeptical without being snide.” A Giants-focused personality named Sal Moreno is framed as a “sports degenerate” whose work avoids “no bro-science, no Rogan-style credulity, no right-coded grift.”
The Dissent is focused on aggregation. But its citation practices are still developing. Its bot reporters tend to mention sources without providing hyperlinks, and Carrasco says, “I’m trying to work on that.”
The open question is what counts as journalism
Northwestern professor Nicholas Diakopoulos, who runs the university’s Computational Journalism Lab, sees these startups as part of an “experimental phase” for media powered by generative AI. He says, “It's not yet clear to me that there's much audience for these AI-agent-written news sites.”
He is also doubtful that mainstream journalists would give AI agents so much control over wording and framing, because those choices affect integrity, legality and accuracy. His research points to another issue: when AI chatbots search for sources, they can surface 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 topics.
Pete Pachal, founder of a newsletter and podcast about generative AI and the media, draws a line around reporting that depends on human trust. “Cultivating the trust of a source, I do think that’s going to be human-only,” he says. But he also sees a role for AI systems in certain kinds of work, such as finding scoops in large datasets or blogging live events like an Apple product launch.
The immediate future of AI newsrooms may therefore be uneven. They can publish quickly, operate cheaply and cover more ground than a small human team. But the harder tests remain accuracy, sourcing, accountability and whether readers will value news written and packaged by agents.