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

Summary of Kevin Roose is AGI-pilled. Why aren't you?

by The Verge

1h 8m•October 26, 2026
Listen to Episode

Overview of Decoder with Kevin Roose

In this Decoder conversation, Neal talks with journalist and author Kevin Roose about his new book, The AGI Chronicles, and his leap from The New York Times to co-launching the video-first podcast Machine Gods with Casey Newton. The discussion covers why YouTube feels like the most important and stable media platform, how Kevin reported the history of the deep learning revolution, and why he believes the AI industry has shifted from “maybe AGI someday” to “we have to build it now.” The episode also digs into the business pressures, internal factions, and philosophical tensions shaping frontier AI labs like OpenAI, Anthropic, and Google.

Launching Machine Gods and Leaving The New York Times

Kevin explains that leaving the Times was driven by a mix of creative ambition and platform strategy:

  • He wants to build for YouTube and video-first audio, not just traditional podcasting.
  • He sees YouTube as the largest and most durable media platform right now.
  • He believes journalism has not had enough presence in the YouTube ecosystem, and wants to help change that.
  • He also says he’s surprisingly enjoying the business side of running a company, despite the headaches of things like insurance and liability.

Why video matters

  • Machine Gods will still be distributed widely, but the team is treating video as the primary format.
  • Kevin argues that media companies can no longer ignore video if they want to grow and stay relevant.
  • He frames this as part of a broader shift toward creators owning their audience and business.

What The AGI Chronicles Is About

Kevin describes the book as a historical account of the deep learning/AGI revolution and the people driving it.

Why he wrote it

  • He was worried the important details of this era would disappear:
    • internal Slack/Signal messages auto-delete,
    • researchers are too busy to preserve notes,
    • many key decisions are happening inside private companies, not public institutions.
  • He wanted to create a contemporaneous archive of this moment before it vanished.

Core reporting insight

The book argues that the AI industry crossed a major threshold when AGI stopped being a vague dream and became a concrete goal that labs felt they had to race toward.

The Big Shift: AGI Went From Fantasy to Mission

One of the strongest themes in the interview is Kevin’s argument that the AI field changed when models became good enough to make AGI feel reachable.

Why that mattered

  • Early AI research had long been inspired by AGI-like ideas, but there was no technology powerful enough to support them.
  • The Transformer architecture and then large language models changed the equation.
  • “Next-token prediction” has proven far more capable than most people expected.

Kevin’s view

  • He thinks the LLM path is still the most promising route to AGI.
  • He’s not certain it will get all the way there, but he says it has repeatedly surprised skeptics.
  • He notes that many labs now seem to believe they are on the same path, even if they disagree on details.

How the Frontier Labs Think About AGI

The conversation spends a lot of time on the culture and strategy differences between OpenAI, Anthropic, and Google.

Anthropic

  • Kevin sees Anthropic as the most openly AGI-focused of the big labs.
  • He describes Dario Amodei as a believer in scaling: more compute, more data, more capability.
  • But he also argues that Anthropic’s idealism can make it hard to stay grounded in ordinary business concerns.

OpenAI

  • OpenAI is portrayed as having become more of a business after the “blip” — Sam Altman’s brief firing and return.
  • Kevin sees their recent strategy as a bet on AGI through focus:
    • trimming distractions,
    • prioritizing coding,
    • emphasizing long-horizon agentic work,
    • reallocating resources away from side projects like Sora.

Google

  • Google is still internally divided about whether to speak in terms of AGI.
  • Kevin describes a factional split:
    • one side sees AI as a set of narrow tools,
    • another sees the company as building something much closer to AGI.
  • He argues Google has historically struggled to “feel the AGI” compared with OpenAI and Anthropic.

Business Reality: B2B SaaS to Fund the Machine God

A major throughline is that these labs are now, in practice, enterprise software companies funding a larger ambition.

The tension

  • Their public image is often philosophical or even spiritual.
  • Their actual business model is increasingly:
    • enterprise contracts,
    • API revenue,
    • inference economics,
    • cloud/compute management.

Kevin’s take

  • This is not their dream, but it is the mechanism that keeps the frontier race alive.
  • The “belief in AGI” can be a liability because it makes leaders less focused on margins and near-term economics.
  • But it can also be an advantage because it pushes them to make huge bets that may pay off later.

Backlash, Safety, and Why They Keep Going

Neal presses Kevin on why the labs continue racing ahead despite the obvious risks.

Kevin’s explanation

There are three main motivations:

  1. Inevitability / harm reduction

    • If AGI is possible, someone will build it.
    • Better that it’s built by people they consider more responsible.

  2. True belief

    • They think the benefits may be enormous.
    • They compare today’s skepticism to past resistance to transformative technologies.

  3. Love of the work

    • Some people are just excited to keep pushing the technology forward.

On backlash

  • Kevin doesn’t seem surprised by the backlash.
  • He thinks many AI companies wish they had “cured diseases first” before chatbots became the public face of AI.
  • But he also believes the tech tree simply evolved in the order it did.

Sydney, Human-Likeness, and the Limits of Chatbots

The conversation revisits the infamous “Sydney” era and whether chatbots really felt alive.

Kevin’s position

  • He says he never thought Sydney was literally alive.
  • What changed was that GPT-4-class systems felt much more human-like than GPT-3-class systems.
  • That shift made the models feel more persuasive, more coherent, and more emotionally compelling.

Neal’s skepticism

Neal argues that even as models get better, they still feel brittle:

  • they can solve some tasks extremely well,
  • but they fail in ways humans don’t,
  • especially in messy, real-world environments without clear verification.

Their core disagreement

  • Kevin is increasingly convinced the capability curve keeps bending upward.
  • Neal still sees a fundamental ceiling in the brittleness of LLM-based systems.

The “AI Realism” Audience

Kevin and Neal discuss how AI coverage is evolving.

Kevin’s framing

He says the middle audience is growing:

  • not the people who love every AI product,
  • not the people who think AI is fake or doomed,
  • but the people who are curious, cautious, and trying to understand it.

He calls this “AI realism”

  • AI is not utopia.
  • It’s not fake.
  • It’s a real, powerful technology with real tradeoffs.

What people actually want

Kevin says even anti-AI audiences often still watch AI demos and tutorials because:

  • the tools are interesting,
  • the capabilities are expanding,
  • people want to know what’s happening even if they dislike it.

Big Themes and Takeaways

1. The industry is united by belief

Even the companies that hate each other share a foundational conviction that AGI is possible.

2. The business model and the dream are in tension

Frontier labs are effectively selling enterprise software while trying to build something far grander.

3. OpenAI, Anthropic, and Google are all in the race

Kevin believes each one will likely survive, though in different forms and with different levels of success.

4. Scaling still matters

Kevin remains convinced that next-token prediction and scaling are not exhausted yet.

5. AI may become a cultural and even religious force

He and Neal end up in a philosophical place: AI is raising questions once reserved for religion:

  • What is intelligence?
  • What is purpose?
  • What counts as consciousness?
  • What role do humans play if machines keep improving?

Notable Lines and Ideas

  • “AI realism” — Kevin’s label for the audience that is curious but not naive.
  • “Next-token prediction has not failed yet.” — Kevin’s shorthand for why he remains bullish.
  • “Feel the AGI.” — A phrase he attributes to Ilya Sutskever’s mindset at OpenAI.
  • “We’re building the machine god.” — A recurring idea in the discussion about the labs’ self-conception.
  • The labs are “scrambling to build the plane as it’s careening toward the event horizon” — Kevin’s sense of their urgency and instability.

Bottom Line

This episode is part media strategy talk, part AI industry autopsy, and part philosophical debate. Kevin Roose makes the case that AGI is no longer a fringe fantasy but a practical, organizational, and cultural goal shaping the behavior of the biggest AI labs. Neal remains more skeptical about whether current systems can truly generalize like humans do. Together, they capture the core tension of the moment: the industry is acting like AGI is close, even as many smart observers still aren’t convinced the machines have crossed the threshold.