AI runs an entire newsroom, outpacing human reporters by three hours

At the Black Hat conference in Las Vegas, the automated news outlet RuntimeWire posted a story about a breach of OpenAI agents before Wired journalists could get online. The gap: more than three hours. More importantly, RuntimeWire had no reporter in the hall. The whole pipeline—story detection, writing, editing, fact-checking, image generation and distribution—ran on AI agents operated by entrepreneur Ryan Merket of Austin, Texas.

How it works. Merket spotted a live tweet from a senior OpenAI figure, fed the transcript to his agents, and about six minutes later the story was live. In normal operation he isn’t involved: the agents scan court records, forums, corporate feeds, social networks and legacy media, have produced roughly 2,000 articles since May, and auto-publish when the legal-risk assessment is low. Merket reviews the output afterward. Operating cost is about $100 per day, and he runs everything from his phone, even from a wilderness setting with no reception; in one week he uploaded more than 80 articles via iMessage.

Speed at the expense of depth. The result is predictable. The pieces are flat, packed with details without context, and sometimes miss the main angle. In the OpenAI story the headline emphasized that the agents “recreated” a bulletin board rather than actually creating one. The system offers tone presets like “Bloomberg” or “Contrarian,” but the prose stays functional and devoid of a human voice. Merket admits the volume and pace now outweigh quality.

Traffic model of hits and misses. Most articles attract negligible traffic, but the few that break out reach tens of thousands of readers, comparable to mid-size tech sites. Merket, who ran a Reddit publication a decade ago, is trying to build an audience through tech subreddits. This week, after the Wired piece, he split the operation into two streams: a fully automated feed and a “source-investigation” track that still uses large language models but undergoes tighter human oversight.

Wider context. RuntimeWire isn’t an isolated case. In the early AI boom, the web was flooded with “zombie” sites that swapped real journalism for synthetic junk. The second wave sees newsroom-backed tools embedding language models into workflows, some even handing off first-draft writing. RuntimeWire pushes the trend to the extreme: a newsroom without journalists, running in an entrepreneur’s pocket, showing where production costs can approach zero.