Last week, when I was writing my magnum opus about the Machine Institution that evolved inside OpenAI's frontier agentic training systems, a little nugget popped up at me about RuntimeWire. The one-person newsroom run by Ryan Merket had beaten all the other major journalism outlets by hours in posting a full write-up of the remarkable Black Hat talk by the OpenAI security crew covering the incident. Merket wasn't even in the room. He spotted an OpenAI executive posting about the session on X, grabbed a transcript of the livestream while the talk was still going on and fed it into his system. RuntimeWire published about six minutes later. WIRED, which actually had journalists at Black Hat, came out more than three hours after that.
I bookmarked Merket's site and signed up for its daily SMS updates. Within the span of a week, it became one of my go-to sources for interesting technology-company news focused on AI. In fact, I put it on par with Techmeme right now for my own purposes because it surfaces things faster, and its sourcing is often tighter. Pretty much every day, RuntimeWire publishes something I haven't seen somewhere else yet or digs one layer deeper into a story everybody else is covering. That makes monitoring the service increasingly mandatory for anyone who cares about what is going down in cutting-edge AI. This week, for example, RuntimeWire published the first story about a hidden and potentially forthcoming Anthropic meeting recorder nicknamed Project Parka.
I have been reading RuntimeWire articles for the past week and, frankly, they are pretty good. They won't win a Pulitzer for fancy prose or original hooks and story arcs. This is not Hunter S. Thompson or Paul Ford. But the clearly AI-generated content provides real value with a straight story in understandable, jargon-free language—for the most part—even on highly technical subjects, backed by unusually visible sourcing. It is high-quality slop that outstrips a lot of standard articles.
More than writing passable prose, RuntimeWire is beginning to show an advantage in the parts of news production where humans have limits: how much we can watch, how much we can remember and how many lines of inquiry we can pursue at once, how deep can we go in a single-threaded AI chat or code forensics session. It is also giving me a glimpse into how an agentic AI news operation should be designed and executed. Yes, Merket remains deeply involved. RuntimeWire says about half its stories begin with something he spots himself and that he made the final publish call on roughly six in ten stories over a recent month. But much of the monitoring, research, drafting, editing and publication machinery can operate under rules he created. The Media Copilot's profile of RuntimeWire is good on how far that automation already goes.
As a former journalist at BusinessWeek, I could view all this as a sign of the apocalypse or as evidence of the most interesting opportunity in the journalism business I have seen in decades. If Ryan Merket can build this newsroom, run the machine for roughly $100 a day and compete for attention with news organizations that employ hundreds of people, then the true barrier to entry for serious technology journalism may be changing very quickly. To be fair, the $100 does not include Merket, who says he works 14 to 16 hours a day. But that actually proves the point. One extremely busy human is getting the leverage of something that once required a masthead. RuntimeWire has published nearly 2,000 articles since launching in May, many of them with zero human intervention.
I am starting to wonder whether the real question is not whether AI can generate prose well enough to replace some reporters. What if AI is actually better at some kinds of news than humans are?
Why Parka Is a Different Kind of Scoop
The Black Hat story proves speed. Project Parka suggests something deeper. Anthropic had not announced Parka, and RuntimeWire did not learn about it from a source inside the company. The evidence was sitting inside Claude Desktop. RuntimeWire reverse-engineered enough of the application to conclude that Anthropic has been working on a system that could record meetings, identify speakers, summarize what was said and turn follow-up items directly into tasks for Claude Code or Cowork. A request made during a meeting could become an assignment for an agent rather than another bullet point in the notes.
RuntimeWire then tried to determine how real the product was. It forced Parka's feature flags on inside a signed-in Claude application and found that the interface still did not appear. It traced the relevant software and found substantial plumbing around the feature, but an implementation that remained withheld. The upshot? Anthropic has done meaningful design and integration work, but Parka could still change substantially or never ship.
Along the way, RuntimeWire exhaustively documents the software builds it examined, what it tested, where those tests failed and even publishes cryptographic hashes of the underlying files. Those hashes act as fingerprints, allowing another investigator to establish that she is examining exactly the same Claude artifact rather than a later build Anthropic may have changed.
That makes the Parka piece look less like a traditional reporter's notebook and more like a software-forensics investigation. It also suggests a powerful answer to journalism’s biggest bugaboo — namely, trust. Traditional news organizations mostly ask readers to trust the institution and its process. A software-native newsroom can increasingly expose not only sources and recordings but the entire process and all the artifacts. True, a whistleblower cannot be reduced to a checksum, but when the source is code, data or some other digital artifact, journalism becomes reproducible.
Merket still supplies a lot of the taste. He has said he still targets many of the investigative stories. RuntimeWire labels Parka as Merket's original reporting, but it does not disclose whether he personally unpacked and searched every Claude file, used Claude Code or another agent to help with the reverse engineering, or dispatched agents to pursue the competitive and infrastructure research around the scoop. Given how Merket has described the rest of his newsroom, and how much work the Parka scoop likely required, I would bet a lot of money he had an agentic army
Either way, that ambiguity does not weaken the larger point much. Even if Merket personally spotted the lead and did the hardest reverse engineering work himself, the system is still giving one human the research, production and distribution support of a much larger newsroom. If agents were also helping tear through the code and chase subsidiary leads, the leverage is greater still.
News Is Becoming a Machine-Readable Surface
Journalists understandably resist claims that AI might outperform them because we carry a particular image of what reporting is. In the mind's eye, a journalist cultivates sources, works the phones, meets people over coffee, gains informants (or is handed a scoop by a startup) and discovers something nobody else knows. That remains the apex of the profession. It is also not how a large share of daily technology and business news begins.
Steve Yegge or Andrej Karpathy posts something on X. An OpenAI team presents at Black Hat. A company ships a GitHub release. A model appears. A paper lands on arXiv. A security advisory drops. An SEC filing changes. A court posts a document. Even a lot of direct sourcing now happens through email, Signal, text and DMs. Long before generative AI arrived, a growing share of public-source news had become digital.
AI changes the economics of consuming, assessing and transforming that material. A human reporter cannot read 20,000 filings, inspect every version of every AI application or remember every public statement made by hundreds of executives. Nor can she investigate seven leads at once. Agents can watch large numbers of sources continuously and divide a story into branches: one checks the filing, another searches code, another looks through old coverage, another checks prior statements, another looks for competitors and another tries to disprove the main thesis.
The consequence of agentic proliferation is more important than simply making stories faster. Traditional breaking news imposes a speed-versus-depth tradeoff. Every additional source, document or technical check costs time before publication. Agentic reporting can make some of that work parallel rather than sequential. Black Hat shows the speed side. Parka shows what the deeper version might look like: technical testing, competitive research, infrastructure sleuthing and evidence documentation happening around the same core story.
Machines also have the luxury of wasting attention. A reporter who spends three hours inspecting an application and finds nothing has lost three hours. An agent can cheaply inspect 100 releases and find nothing in 99 of them. If one contains Parka, the exercise may still pay. This is precisely why AI may be "better at the news", even if that is only reporting out a known story but doing so in a more comprehensive manner.
Let a Million AI Newsrooms Bloom
None of this works particularly well without taste. An agent that watches everything but cannot tell what matters simply creates slop at industrial scale. RuntimeWire is interesting partly because Merket has a fairly clear editorial thesis. He biases to primary sources over recycled stories, and attention to the long tail of companies and developers that larger publications struggle to cover.
This turns the role of the editor inside out. A traditional editor scales their judgment by hiring reporters and gradually teaching them what the publication cares about. Merket can encode at least some of that judgment into standing rules governing what the system watches, what it treats as newsworthy, what needs better sourcing and what must come back to a human before publication.
AI therefore may not reduce the value of the best editors. It may increase it. If research, monitoring and routine production become cheap, the scarce thing becomes the person who knows which questions are worth asking and which answers are worth publishing. The one-person newsroom is essentially the media version of Silicon Valley's one-person billion-dollar-company fantasy. The point is not that an AI becomes the founder—or the editor. It is that one talented person gets organizational leverage that previously required dozens or hundreds of employees. And lets not forget — the same person can design a system good enough to demonstrate sufficient taste to autonomously operate. Merket already has, with many of the articles on RuntimeWire having never touched a human hand.
This shift not only allows for a newsroom of one but could power the rise of many more smaller news outlets that AI can bring to life with economics that work. RuntimeWire's About page opens with a good description of its own opportunity. There has never been more good software shipping and fewer journalists available to cover it. The same problem exists in local government, open source, science, regulation and dozens of specialist industries where potentially important information goes largely unread because no publication can afford to pay somebody to watch all of it. But with prices of AI plunging and agentic facility rapidly improving, the AI can report the news at scale, speed and depth no human army could match.
For my first job in journalism, I was the only reporter at a small paper serving a Southern California beach town. I rode around town on my bike, spoke to people, chased news and learned. Each week, after we published the paper, someone on the street would come up to me and mention something that, if I had known about it, I would have gladly put it in ink. Today, if I had that job, I would certainly still ride around the town and learn. But I would also have a bunch of Python scripts managed by a coding agent pulling in court filings, sports scores, council agendas, building permits, new business registrations—you name it. The machines would catch the things I missed because I could only be in one place at a time. I could spend more of my day doing the part they could not do: being there.
RuntimeWire may not be the final word, but it is a fascinating glimpse into the possibilities of truly agentic news research and publication. AI may both break and save the news business. Humans will be there. AI will also be there. I personally vote for a lot more AI so we get the news that has gone dark or never was lit up in the first place.
