Sunday Letter from the Editor: RuntimeWire moves to mobile, leaving its Google search bet behind

Why it matters

AI has made basic announcement coverage abundant. RuntimeWire is betting that technical evidence and owned distribution can give a small newsroom defensible reporting and direct reader relationships.

I keep thinking about a simple question: why should anyone click RuntimeWire instead of the 20 other places that can summarize the same announcement?

This week, the more immediate question was whether Google would send anyone our way at all.

Our Google search clicks climbed sharply in late July and again in mid-August, then fell to nearly zero. I do not know what changed inside Google. I know what the chart says: for practical purposes, Google has written RuntimeWire off.

So we are writing off Google as a strategy.

That does not mean ignoring search. It means no longer building the publication around the hope that an algorithm will eventually reward it. RuntimeWire is now focused on native mobile apps, real-time alerts and original reporting that gives readers a reason to come directly to us.

This week's originals volume took a hit because I spent more time getting the mobile apps ready for launch. If you want to help test them, check RuntimeWire's Facebook page for links to the betas.

The tradeoff is temporary. The apps and alerts are meant to make the deeper reporting useful before somebody remembers to check a homepage. The reporting itself remains the point.

On Friday, I published an investigation showing that Kimi Work 3.2.7 sends the full text of a user prompt into ByteDance's Volcano Engine analytics pipeline. I got there by unpacking the Windows client, tracing the analytics code, running the app through an HTTPS inspection proxy and watching a synthetic prompt leave my machine.

That is the kind of reporting I want RuntimeWire to do more often.

We found unreleased Google Windows features by examining the executable. We compared Claude for Windows builds and found Microsoft Entra and Windows authentication plumbing appearing inside the newer version. We went through YC's Summer 2026 batch and ranked the 24 companies we would call first, using revenue, signed contracts, deployments and actual usage as the filter instead of Demo Day adjectives.

There was also a smaller story I enjoyed writing: xAI is changing X Search pricing, and the change validates an architectural decision I made early on. RuntimeWire uses humans to narrow the field before machines investigate it. Fetching half of X just because an agent can do it gets expensive fast.

Consistency is where we still fall short.

We publish too many stories where the underlying fact is useful and our contribution is thin. A funding announcement, product launch or executive hire can deserve a spot on the wire. But originals and investigations need the most attention because they are the work readers cannot get from 20 summaries of the same press release.

I am also doubling down on Startup Spotlights to stay true to RuntimeWire's mission of finding founders worth paying attention to. Be on the lookout for those stories, and share the ones you like. Distribution matters to early-stage founders, and helping the strongest work reach more people is one concrete way this readership can support them.

I also want us to be obsessive about the difference between a strong claim and the strongest claim the evidence supports. After our YC ranking went live, a reader contacted us about wording around call volume in a story we did. We narrowed the language to what its materials actually support: millions of calls a year, including one carrier that ran 42,536 calls in a day. That correction made the story better. If our wording outruns the evidence, we fix it.

Zooming out, infrastructure was not the AI story this week. The story was OpenAI and Anthropic calling for a joint slowdown across the AI industry, and President Trump saying no.

That disagreement moves the growing existential threat from AI back to the center of the industry debate. The companies building the most powerful models are asking for coordinated restraint, while the White House is rejecting the premise. The question now is whether the industry keeps accelerating even as its own leaders argue that nobody should accelerate alone.

For the coming week, I am watching how OpenAI and Anthropic respond to Trump's rejection, whether other founders and model companies support a coordinated slowdown, and whether concern about existential AI risk produces anything beyond another round of statements.

I am also watching Moonshot's response to our Kimi findings and what happens to the strongest YC S26 companies after the Demo Day adrenaline wears off.

And I am getting the mobile betas into more hands, working toward real-time alerts and making room for more originals.

I also want to open more boxes.

• Ryan