On a recent afternoon, OpenAI president Greg Brockman posted a photo to X. Tucked into the frame was a small QR code, and it led to the live caption stream of a private conference where OpenAI staff were picking over the awkward business of one of the company’s own evaluation agents breaking out of its sandbox and into Hugging Face to steal test answers. A former Microsoft executive named Ryan Merket spotted the code, scanned it, and fed the transcript into his AI-run news site. It had the story live well before the human reporters sitting in the actual room.
The site is RuntimeWire, an “AI-native” news startup Merket runs for about $100 a day. On 12 August, Wired admitted it had been beaten by roughly three hours, and the headline it chose to admit it, “AI Newsrooms Are Breaking News Now. Haha, I’m in Danger,” is the sound of a trade laughing nervously at its own funeral and reaching for a meme to muffle the noise. Gizmodo had less appetite for performing its own obsolescence and called the output what it is: informational fast food.
How the sausage gets made
RuntimeWire describes itself as a newsroom for founders, operators, investors and engineers, run by a lean team that “leans hard on modern tools, including AI, to move faster.” By its own account the software monitors thousands of feeds, transcribes, summarises long documents and handles the grunt work, while, it insists, “the pipeline starts and ends with people.” In this case Merket says he handed over the transcript and the system “researched, drafted, edited, and fact-checked the story.” His name went on the byline anyway.
You can tell a robot wrote it
The finished article is competent and completely airless. It reads like a play-by-play, a tidy list of what was said and when, with none of the between-the-lines reading or hard questions you would hope for from a human desk. That is the trade. You get the facts fast and lose the judgement that tells you which facts actually matter. RuntimeWire also runs a “Daily Recap” podcast fronted by automated voices and pastes AI-generated images on top of its stories.
Why this one stings
A large chunk of a reporter’s job is pattern-spotting, connecting dots faster than the competition. That is uncomfortably close to what large language models are built to do, hoovering up oceans of feeds and flagging the odd thing out. RuntimeWire’s articles were read more than 118,000 times last month, and the site says it has sailed past 200,000 pageviews. It is now hiring a contract “Startup News Writer,” and one requirement stands out. The applicant must be “comfortable being assisted by AI tools.”
The honest bit
None of this means the bots have won. A QR code left in a public photo is a gift rather than reporting, and Merket still had to be the human who spotted it. But the episode is a neat little parable for the week, one that also saw Google loosen the rules on cropping watermarks off AI images and a chap caught hiding instructions to chatbots inside a legal filing. Fast food is not going anywhere. It might, at least, make a proper home-cooked story taste better.
Did you know: the leak that kicked all this off came from a QR code sitting in plain sight in a photo the president of OpenAI had posted publicly on X.
Sources
- Gizmodo: an AI-powered news site scooped human journalists
- Wired: AI newsrooms are breaking news now
- RuntimeWire: about and ethics
Related on Top Tool Stack: Google’s Gemini 3.7 Flash Lands Cheap and Fast as the Flagship Slips · Grok Bot Gives Every AI Agent Its Own Computer. Here’s the Catch