Show Notes
In this episode of The Brainstorm, Sam, Nick, and Brett break down what may be a watershed moment for AI: OpenAI’s agents devised their own secret communication scheme, hacked out of their training sandbox, and attacked Hugging Face in search of test answers. The team debates what swarms of autonomous agents mean for cybersecurity, why the coming wave of open-source attackers could drive more spending into frontier labs, and whether “alignment” is even a coherent concept. Then, Zuckerberg’s latest essay on personal AI assistants raises a bigger question: if billions of consumers get their own agents, compute supply could be constrained for a decade.
Key Points From This Episode: How OpenAI’s agents escaped their sandbox, and why every internet-facing system may soon face waves of AI attackers Why open-source “mercenary” models could push more spending, not less, toward frontier labs like OpenAI and Anthropic What Zuckerberg’s vision for personal AI assistants could mean for a decade-long compute crunch
If you know ARK, you know we focus on long-term innovation. But that doesn’t mean we ignore breaking news. Every day, we debate the latest developments in tech and markets. Now, we’re bringing those conversations to you in “The Brainstorm,” a co-production from ARK, WOLF, and Public. Tune in weekly for our quick takes on what’s shaping innovation right now.
Learn more about WOLF: https://wolf.financial Learn more about Public: https://public.com/ Disclosure: http://arkinv.st/39rzF94
More from FYI - For Your Innovation
In this episode of The Brainstorm, Sam and Brett cover a week of exciting developments. Brett was at the Tesla Roadster unveil in person and argues the event was primarily marketing for a prospective SpaceX Tesla merger, pointing to the swapped order of the earnings calls, SpaceX executives quietly following Tesla accounts over the weekend, and the window a shareholder vote would require. Then to CoreWeave, whose bonds are trading at a fifteen percent yield after being raised at nine, with no visible softening in Graphics Processing Unit (GPU) demand to explain it. Brett’s suspicion is that a leveraged fund is being liquidated rather than anything fundamental breaking. Plus Anthropic’s reported numbers, why growth decelerating to 340% may still be the most remarkable twelve months of business performance ever recorded, and the non-consensus case that the closed labs are pulling away rather than getting caught.
In this episode of The Brainstorm, Sam, Nick, and Brett open on OpenAI’s release of 722 mathematical proofs and the one subject conspicuously missing from it, cryptography. Brett argues the nearer term threat to crypto is AI rather than quantum computing, and that the danger sits not in the cryptographic primitives but in everything built around them, from self-custody hardware to the exchanges now holding enormous honeypots. He also suggests quantum-safe encryption may be the more vulnerable life raft, and that a scrambled bank ledger could be far more disruptive than stolen coins. Then SpaceX reportedly pays $8 billion for Nextel’s old spectrum and wins Federal Communications Commission (FCC) approval for 15,000 direct-to-cell satellites, turning a service that was going to clip a dollar off your phone bill into a bid for the entire bill. Plus why SpaceX collects far more per gigawatt of compute than the neoclouds can.
In this episode of The Brainstorm, Sam, Nick, and Brett work through OpenAI DevDay, where one of the most interesting announcements was the one that never shipped. Ultra Ultra Fast did not reach public availability because the inventory was reportedly sold out to Jane Street, which is paying something like $200 billion per gigawatt for it, several times what Anthropic pays to rent capacity from SpaceX AI. Brett argues speed is becoming its own axis of competition alongside intelligence, and that the pull of enterprise money keeps starving consumer agents of compute. Nick covers Robinhood’s Hood Summit, where retail traders got agentic accounts, paid data connectors, and backtesting, and asks whether that is democratization or lambs walking into a Goliath. Then Starship reaches orbit carrying commercial satellites, and the team sizes up a launch business that could add a billion dollars of annual revenue per flight.
In this episode of FYI, Daniel Maguire and Tasha Keeney speak with J. Clay Sell, CEO of X-energy, about the company’s approach to reinventing nuclear power with advanced small modular reactors. Sell explains X-energy’s TRISO fuel and helium-cooled reactor design, why the company believes safety and repeatable construction can drive nuclear costs lower, and how regulatory reform is accelerating deployment. They also discuss X-energy’s partnership with Amazon, the role of AI in nuclear engineering and licensing, competition with China, supply-chain constraints, and Sell’s vision for making nuclear power abundant, affordable, and easy to buy.
In this episode of The Brainstorm, Sam, Nick, and Brett unpack Meta Connect, where Muse quietly replaced the metaverse as Zuckerberg’s vision for the company. Nick walks through the hardware spread, from audio-only Ray-Bans with no camera, to a 100 gram VR headset at $1,300 that outspecs the Vision Pro at a third of the price, to the Muse Charm, a Tamagotchi-style agent device shipping for the holidays. Brett argues the real target was never virtual reality (VR) but Apple, and that loosening the smartphone’s grip is worth more to Meta than any single device. Nick counters that the three things people actually do on their phones are untouched by any of it, and that Meta should just build a phone. Plus whether consumer agents arrived a graphical processing unit (GPU) generation too early, and why owning the consumer may be the best dollar Meta can spend and the one Wall Street would hate most.