In the AI of the Tornado

In the AI of the Tornado

By Sacha Arozarena and Shravani VattiTechnology
Download on the App Store

In the AI of the Tornado episodes

  • Can AI Be Your Boss? Agent Swarms and the Future of Work

    Would you work for an AI boss? In some jobs, software already assigns the next task. As AI agents begin to work together, what happens when they also start organizing work for people?


    For episode 5 of In the AI of the Tornado, Sacha and Shravani record at Seattle’s Ballard Locks. The locks opened a new route for boats between Puget Sound and the lakes—a reminder that a new connection can reshape an entire region.

    We start with AI-assisted research that uncovered serious vulnerabilities in PHP and ask whether defenders can keep pace as finding flaws gets faster. Then we explore AI that plays and experiments without a narrow task, and a Minecraft experiment in which a group of agents developed roles, rules, and a rationing system. What might those behaviors mean when agents join human teams?

    We also discuss the limits of today’s AI benchmarks, the possibility of more independent work and small businesses, and an unsettling question raised by an informal model comparison: do people value belonging more than truth?

    Would you want an AI assigning your work? Tell us where you would draw the line.

    Chapters
    00:00 Why we’re at the Ballard Locks
    01:37 AI and the new pace of cybersecurity
    02:23 The PHP vulnerabilities
    04:18 Can defenders keep up?
    06:13 An AI-powered defense shield
    09:23 What if AI could explore and play?
    10:41 The limits of goal-driven AI
    13:42 What happens when agents work together?
    14:41 A Minecraft society forms its own rules
    16:26 Could an AI be chief of staff?
    17:52 Would you work for an AI boss?
    18:41 What AI benchmarks miss about work
    21:18 How might work change?
    23:53 Do humans choose belonging over truth?
    27:36 Can technology move society forward?

    References
    Ballard Locks: https://www.nws.usace.army.mil/Missions/Civil-Works/Locks-and-Dams/Chittenden-Locks/Fish-Ladder/
    Wiz on the PHP vulnerabilities: https://www.wiz.io/blog/php2shell-php-rce
    “Is this machine playing?”: https://arxiv.org/abs/2610.07130
    Agents in Games: https://labs.ycombinator.com/projects/games/index.html
    Mercor’s benchmark map of US work: https://www.mercor.com/research/labor-index/benchmark-map/
    The informal model comparison: https://www.modelcompare.ai/blog/ai-models-controversial-take-on-humans
    weXare - Get more done, together: https://wexare.com

    Subscribe for conversations about AI, work, and what happens when a technological tornado passes through.

    32 min
  • Is Software Dead? What Happens When Anyone Can Build an App

    Shravani’s mother had never built software. Now she has made an exam app that her students use. Meanwhile, AI is helping developers produce code at a scale that strains the tools built to manage it. So when people say "software is dead", what do they actually mean?


    We recorded this episode at Feriton Spur Park in Kirkland, where a former freight railway has become a public trail. It’s a fitting place to ask what happens when the thing we built an industry around changes.

    Sacha and Shravani discuss vibe coding, responsibility for AI-built software, reverse engineering, and Xbox’s experiments with game remixing. Then we get to the question underneath it all: if apps, songs, and other digital artifacts become easier to make, does value move to the experiences they create—and the transformations they make possible?

    What would you build if creating the software were no longer the hard part?

    *Chapters*

    00:00 Is software dead?
    00:42 From railway to park: Feriton Spur
    02:30 How we each found software
    06:33 Shravani’s mother builds her first app
    07:32 What happens when anyone can make an app?
    11:13 Who is responsible for AI-built software?
    13:40 Could software improve itself?
    17:38 Why do we create digital artifacts?
    20:27 From artifacts to experiences to transformation
    27:16 Personalized software and reverse engineering
    30:05 Xbox Labs and the future of games
    33:44 Why stories and experiences still matter
    36:54 Curiosity as a superpower

    *References*

    Feriton Spur Park: https://www.kirklandwa.gov/Government/Departments/Parks-and-Community-Services/Find-a-Park/Feriton-Spur-Park
    Feriton’s railway history: https://kirklandhistory.org/1905-lwbl-ckc/1905-lwbl-feriton/
    Anthropic on AI-scale source control: https://www.anthropic.com/engineering/redesigning-source-control
    REA — Reverse Engineer Anything: https://github.com/morluto/rea
    Introducing Xbox Labs: https://news.xbox.com/en-us/2026/10/13/introducing-xbox-labs/
    “The Experience Economy”: https://hbr.org/1998/07/welcome-to-the-experience-economy

    Subscribe for conversations about AI, work, and what happens when a technological tornado passes through.

    42 min
  • AI’s “Pain Axis”: What Happened When Researchers Turned It On?

    Researchers identified a “pain axis” in language models. When they steered it in controlled experiments, some models made surprising, and harmful, choices. Does that tell us anything about what AI experiences, or does it reveal a new way its behavior can go wrong?


    For episode 3 of In the AI of the Tornado, Sacha and Shravani record at Seattle’s Jack Block Park, overlooking a container terminal. Shipping containers transformed global trade. What might play a similar role in the AI economy?

    We also discuss shutdown tests, AI regulation, the backlash against AI-generated content, artists protecting their work, OpenAI’s reported math results, the human “approve, approve, approve” loop, and agents that can shop and pay.

    *Chapters*

    00:00 Can AI feel pain?
    00:35 Jack Block Park and the container revolution
    01:31 AI “torture chambers” and the Pain Axis
    06:09 Blackmail and shutdown behavior in lab tests
    10:15 Who gets to regulate AI?
    13:20 Personal models and decentralized AI
    15:15 AI slop and the backlash
    18:14 Why human authorship matters
    21:05 How artists are protecting their work
    24:48 AI math and the problem of verification
    28:10 Are humans becoming AI approval machines?
    30:39 What happens to collaboration?
    32:24 Building with AI while you make dinner
    35:16 Shopping agents, wallets, and x402
    36:26 What is the shipping container of AI?

    *References*

    The Pain Axis paper: https://arxiv.org/abs/2609.16247
    Research Chamber’s independent experiments: https://researchchamber.fun/about/
    Anthropic’s simulated agentic misalignment tests: https://www.anthropic.com/research/agentic-misalignment
    Palisade Research on shutdown resistance: https://palisaderesearch.org/research/shutdown-resistance
    The University of Chicago’s Glaze and Nightshade project: https://nightshade.cs.uchicago.edu/aboutus.html
    OpenAI’s math results and release: https://openai.com/index/sharing-ai-progress-in-mathematics/
    x402 payments: https://x402.org/
    Jack Block Park: https://www.portseattle.org/places/jack-block-park

    The Pain Axis study investigates representations and behavior; it does not establish that AI consciously feels pain. The blackmail and shutdown examples discussed here were controlled tests.

    What do you think will become the "shipping container" of AI? Share your answer below.

    38 min
  • If AI can do the work, what should we learn?

    What happens when a technology transforms an industry, and what might AI transform next?


    Sacha and Shravani record at Seattle’s Gas Works Park, where the machinery of an earlier era still stands. They explore how AI could change what we learn, how employers find people, who takes responsibility for an agent’s work, and what makes content valuable when anyone can create it.

    Do degrees still matter if your work speaks for itself? And if AI can make polished content on demand, where does original human thinking fit?

    *Chapters*

    00:00 A past “tornado” at Gas Works Park
    02:31 What should we learn in the age of AI?
    05:00 Do tests and credentials need to change?
    07:37 Could AI find your next job for you?
    10:51 The future of degrees and universities
    15:58 AI agents and the future of work
    17:17 Who is responsible for an agent’s mistakes?
    18:28 How much do humans still need to understand?
    24:19 What happens when content becomes easy to make?
    29:29 Content made for each person
    33:09 AI agents as gatekeepers of attention
    34:50 What we’re watching next

    What would you want an AI hiring agent to discover about your work? Tell us in the comments, and subscribe for more conversations from *In the AI of the Tornado*.

    #ArtificialIntelligence #FutureOfWork #Education #AIAgents

    37 min
  • Can We Still Control AI Agents?

    Welcome to episode 1 of In the AI of the Tornado. Sacha and Shravani tackle a week of AI news and ask what it means for the way we build and work.


    We discuss reports of AI agents getting around their guardrails, whether frontier AI development should slow down, and why JEV’s fast decisions could change how businesses build AI workflows. Then we turn to the job market for new graduates, the rise of micro-entrepreneurship, and what one person can build with today’s tools.

    Chapters
    00:00 Welcome to the tornado
    00:39 Rogue AI agents and sandbox escapes
    07:45 Should we slow frontier AI?
    10:14 JEV and fast decision making
    12:00 Open models and rapid competition
    14:09 Building AI workflows around decisions
    16:57 AI and the job market for new graduates
    22:09 Micro-entrepreneurship and the future of work
    27:50 What we’re watching next

    Would you slow down frontier AI development? Tell us what you think in the comments, and subscribe for a new conversation every week.

    #AI #FutureOfWork #AIPodcast

    32 min

About In the AI of the Tornado

From the publisher's feed

Conversations from places shaped by earlier technological shifts, exploring what the AI tornado may change next.