RuntimeWire says its AI newsroom published 986 stories in 30 days

Why it matters

RuntimeWire's figures document the scale of one founder's automated newsroom.

Ryan Merket, a two-time startup founder whose companies Ping.fm and Appbistro were acquired, says his AI-powered publication RuntimeWire published 986 articles in the 30 days before October 3rd. Its median time from a story being picked up to publication was 36 minutes.

Those are figures Merket reported from RuntimeWire's production database in an October 3rd blog post. The post has more on the operation's output, software and infrastructure. The figures are company-reported, not independently audited newsroom-performance measures.

The 30-day newsroom figures

• 986 articles published, or about 33 a day.
• 36 minutes median turnaround, from story pickup to publication.
• 6.1 cited sources per article, on average.
• 77 monitored feeds: 49 RSS feeds, 21 company newsrooms and seven X accounts.
• 886 stories screened by a curator in the 24 hours before the post. That's an intake figure, not a publication count.

The feed count describes what the system watches, not the breadth or independence of the evidence in each story. The post does not explain how RuntimeWire defines a story as "picked up," calculates the median or counts a cited source. Those definitions would help readers interpret the figures, especially at this publication volume.

Models, media and code

Merket's other figures describe the software behind the output:

• 240,231 AI model calls since RuntimeWire began tracking them on August 4th, including 5,150 in the preceding 24 hours.
• 1.8 billion tokens processed in total, including 957 million in the last 30 days.
• 157 models from 12 providers used during the month.
• 35,962 generated images, videos and audio clips, 289 AI-model comparisons scored by judge panels and 121 video recaps.

The post also lists 636,000 lines of TypeScript across 2,731 files, 669 automated test files, 353 database tables, 785 API endpoints and 2,853 commits. Those are company-reported codebase counts, not measures of reliability or editorial quality. In a May account of building RuntimeWire, Merket described a TypeScript-based publishing system that handles story selection, drafting, checks and distribution. His stated bet was that automation could let one person publish around the clock.

Merket remains RuntimeWire's sole human employee. The RuntimeWire site says the publication has run up 3,706 articles since May 15th. Its operating model puts software in charge of work that a newsroom would otherwise divide among multiple people.

The efficiency case has drawn attention beyond Merket's own posts. In August, WIRED reported that RuntimeWire beat its coverage of an OpenAI security presentation at Black Hat by more than three hours. Merket told WIRED he got the presentation transcript to his agents and publication took about six minutes from that point. The same WIRED report described a system that can publish some stories without Merket's prepublication review when its AI editor assigns them low legal risk. It also criticized the outlet's writing and framing, and reported three corrections at that time.

Merket later defended his Black Hat framing in a follow-up, arguing that the story's key detail was that the agents rebuilt their coordination channel using filenames after OpenAI shut down the original message board. He said agent communication through message boards was already familiar from Moltbook; the reconstruction was the new behavior his article emphasized.

In an August 16th LinkedIn audit, Merket compared RuntimeWire with 11 other technology newsrooms over a 15-day period and ranked his publication third overall, with a score of 4.5 out of 5. He said RuntimeWire published 345 eligible stories in that window, or 23 a day, and described its strongest work as original investigations and product reverse engineering. The ranking and assessment are Merket's own, rather than an independent newsroom audit.

Before building RuntimeWire, Merket worked as an early product manager at Reddit and on Facebook's early platform team. He also co-founded Ping.fm and Appbistro, both acquired. In his May explanation of RuntimeWire, Merket described finding stories and packaging them for multiple channels as an engineering problem. The October figures show how the company tracks that work through model calls, source feeds, articles, code and distribution products.

The update also lists a native iOS app, an encrypted chat app called INSiDE and an API server alongside the website. It does not provide usage or revenue figures for those products, or traffic and revenue figures for RuntimeWire overall. Merket says more on traffic and revenue will be announced later. RuntimeWire has separately described an MCP-based advertising workflow that lets AI agents check fixed-price placements and submit bookings, with human review before an ad runs. The workflow suggests a possible business path for an operation whose core public metric so far is how much it can publish.