AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Kevin Roose is AGI-pilled. Why aren’t you?

Collected Oct 26, 2026

Kevin Roose, the journalist and author of the new book The AGI Chronicles, appeared on Decoder to discuss leaving The New York Times and launching Machine Gods, a tech podcast he co-hosts with Casey Newton. The show launched last week and is being made in partnership with NPR. Roose and Newton previously worked together on Hard Fork, the popular Times podcast, and their departure from the paper marks a move into independent, creator-owned media.

The new venture has a corporate structure, of a sort. Roose said he holds the CEO title after winning a coin flip, a role he described with characteristic dryness as existing mainly so there is someone to put in front of Congress when a scandal breaks. He said the administrative side of running a company is new to him, and that he has been surprised by how much he enjoys the business side, though negotiating insurance quotes for defamation coverage is not the part he finds compelling. Newton, by contrast, has run a business before.

Machine Gods is intended to be video-first. Roose said the team will care about how the show looks to a degree that past audio work did not require, while still making something worth listening to for people who prefer podcasts without video. Distribution will not be exclusive to YouTube: the show is also putting video on Apple Podcasts and Spotify, and Roose said they intend to be everywhere. His stated reasoning is that YouTube is currently the most important media platform and the most stable place to build, even as it changes, and that high-quality journalism occupies a very small share of it. He cited The Verge, Joanna Stern and MKBHD as examples of what is possible there, and said he hopes to bring other people into the operation and settle on formats rather than build everything around himself.

On the audience for AI coverage, Roose described a shift. He said there was for a while strong demand for reassurance that the technology was not happening or did not matter, and that this strain persists in parts of YouTube and on Reddit, which he characterized as very anti-AI. But he believes most people, or at least a plurality, are now either resigned to or excited about AI not going away, and that Machine Gods is aimed at that group. He framed the editorial stance as neither utopian nor dismissive: not telling viewers the technology is wonderful, and not telling them it is fake and will fade without consequences. He credited Newton with the phrase "AI realism" for the position of people who are curious, interested and perhaps a little scared, and whose response to fear is to understand the subject rather than avoid it. He pointed to a practical signal from Hard Fork: a half-hour Claude Code walkthrough in which the hosts built a website on camera performed well, even when the comments came largely from AI skeptics.

The book grew out of a concern about preservation. Roose, who describes himself as a history buff and a reader of books about the Wright brothers, rural electrification and the PC revolution, noted that such histories rest on contemporaneous evidence such as lab notes and correspondence that was archived. He concluded that little comparable material would survive for this wave of AI, for three reasons he laid out: the builders work at private companies rather than government, so they are not legally obliged to preserve records; lawyers configure Signal and Slack messages to auto-delete after set periods; and researchers are too busy to keep notes. He described the book as an effort, partly in the role of a historian, to capture those stories while the people involved were still willing to talk, much of it on background so sources could be protected from the consequences of speaking to him.

On the technology itself, Roose traced the concept of AGI back to antiquity and especially to the 1950s, when the term artificial intelligence was coined and Alan Turing wrote about a thinking machine capable of anything the human brain could do. The idea existed for decades without a technology that could deliver on it, he said; what had to arrive was the transformer, the architecture Google developed and published in 2017 and presented to, among others, OpenAI, which built on it — the T in ChatGPT. He noted the irony that Google invented the approach but did not turn it into a product before OpenAI did. In his account, large language models gave the old dream something to attach to: a technique that appeared able to scale and keep improving across orders of magnitude, which set off the realization at OpenAI and later at Google that the tool for the long-sought goal finally existed. He also observed that next-token prediction has not yet failed at the tasks people want, even though substantial scaffolding — agent harnesses, reinforcement learning and more — has been built around it.

This suggests the more interesting question for technical readers is not whether scaling continues but what share of capability comes from the core model versus the systems wrapped around it. Roose's account of harnesses and reinforcement learning implies that progress is increasingly a product of engineering around the model, which changes where the difficult work sits for teams building on top of these systems. His preservation argument also has a practical edge: if internal messages are set to auto-delete and researchers keep no notes, the institutional memory of how these systems were built is being actively discarded, and that loss will be felt by anyone later trying to reproduce results or understand design choices.

Why it matters: for developers and product teams, the interview is less a technical roadmap than a signal about how the AI story will be covered and consumed. A video-first, independent outlet funded partly through an NPR partnership and distributed across YouTube, Apple Podcasts and Spotify points to a maturing audience that wants evaluation rather than either hype or denial — which, by inference, raises the bar for how products are demonstrated and explained. Roose's emphasis on scattered, disappearing records is also a reminder that the documentation practices teams choose now will determine how legible their work is to everyone later.

Read at The Verge · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Today, I’m talking to Kevin Roose, journalist and author of the new book The AGI Chronicles, which details the recent and explosive transformation of the AI industry. Kevin was a longtime New York Times reporter and host of the popular podcast Hard Fork, but he and friend of the show Casey Newton just left the […]