
Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Sam Altman’s testimony turned the OpenAI trial into a credibility fight — Musk’s lawyer openly pressed Altman with “Are you completely trustworthy?” while Altman responded in a notably softer tone than Musk and framed OpenAI’s for-profit evolution as necessary because, as Ilya Sutskever put it, “If there’s no funding, there’s no big computer.”
The case keeps swinging between nonprofit ideals and power politics — testimony and old emails suggested both Altman and Musk were deeply worried about Google in 2015, while Brockman and others argued Musk himself had entertained folding OpenAI into Tesla and supporting a for-profit structure if he controlled it.
Anthropic secondary-share mania is colliding with securities law and transfer restrictions — the hosts zeroed in on a now-deleted post bragging that brokering an Anthropic secondary deal made more than a decade of prior net worth, then unpacked why direct brokering likely requires a broker-dealer license and why Anthropic is publicly warning that unauthorized transfers are “void.”
The SPV boom is getting absurd enough to feel systemic — the show described retail-style frenzy around private AI equity, including people allegedly selling houses for Anthropic exposure and “taxi cab driver” levels of SPV chatter, with the real risk being messy lawsuits if sellers later try to unwind deals after huge valuation jumps.
Mira Murati’s new startup is chasing real-time multimodal AI, not chatbot glue code — Thinking Machines Lab showed “interaction models” built for full-duplex audio and video so the system can listen and respond simultaneously, which the hosts tied to the near-future promise of live translation tools that feel more like sci-fi hardware than today’s laggy voice assistants.
The back half of the show veered into internet-business theater — eBay quickly rejected GameStop’s $55 billion cash-and-stock bid as “neither credible nor attractive,” while Byron Allen agreed to put $120 million into BuzzFeed for a 52% stake, prompting genuine confusion from the hosts about what valuable strategic asset is left in the BuzzFeed brand.
The episode opens with the OpenAI v. Elon Musk trial, which may wrap earlier than expected, and the hosts frame the stakes clearly: Musk says he was manipulated into funding a nonprofit that later morphed into a for-profit powerhouse. Recent testimony from Mira Murati, Helen Toner, Siobhan Zilis, Satya Nadella, and Ilya Sutskever had already turned the case into a messy referendum on Sam Altman’s candor, the board’s competence, and Microsoft’s role as the “stabilizing partner.”
One line from Ilya steals the segment: “If there’s no funding, there’s no big computer,” which the hosts immediately treat as an all-timer because it compresses the whole OpenAI transition into one brutally simple sentence. They also note the awkward detail that Sutskever’s stake may be worth around $7 billion, which doesn’t exactly make his motivations look pristine to a judge or jury.
As Mike Isaac live-posts from court, Altman’s side tries to show he and Musk were both longtime AI obsessives who feared Google would get there first. Then the stranger details roll in: an old Tesla discussion where OpenAI might have been folded into Tesla, Altman saying Musk implied he’d do the AI work there anyway, and a surreal courtroom image of Altman recounting Elon showing “memes on his phone,” which the court reporter apparently had to hear again out loud.
Musk’s lawyer, Steven Molo, wastes no time and asks the question everyone knew was coming: “Are you completely trustworthy?” The hosts say the contrast with Musk’s own testimony is striking — Altman sounds humbled, careful, almost painfully self-aware, admitting “misunderstandings and a breakdown of trust” while insisting he never intended to deceive the board during the 2023 “blip.”
The show then pivots into the AI secondary-market gold rush, sparked by a deleted post boasting that arranging an Anthropic secondary deal made more money than the poster’s entire net worth from their 20s. That kicks off a lively explanation of why straightforward securities brokering generally needs a broker-dealer license, and why some players instead use SPVs to “facilitate” exposure without technically brokering the shares.
Anthropic’s response was public and sharp: unauthorized stock sales or transfers not approved by the board are “void” and won’t be recognized on the company’s books. The hosts make the real point here: this isn’t a new rule, just a company loudly reminding everyone that clever side deals, futures-like contracts, and SPV wrappers may still blow up later — especially if a seller wants to claw back upside after the stock appreciates massively.
In one of the funnier beats, they use a Frog and Toad analogy from Paradigm’s Frankie to explain SPV mechanics like they’re reading to preschoolers. Beneath the joke is real anxiety: the hosts compare the spread of Anthropic and OpenAI SPVs to a speculative fever, citing stories like people selling houses for Anthropic secondary and warning that unwinding these structures could become its own billion-dollar legal industry.
From there, the mood lightens with Mira Murati’s Thinking Machines Lab unveiling “interaction models” for full-duplex audio and video — AI that can process and respond in real time instead of waiting for turn-taking. The hosts riff on translation masks that sound like a Bane prop from sci-fi, then close on two business oddities: eBay instantly rejecting GameStop’s $55 billion bid, and Byron Allen paying $120 million for a controlling 52% stake in BuzzFeed, leaving everyone wondering what the actual strategy is behind buying a brand now dominated by endless celebrity listicles.
Share
Keep Reading
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.

Playbook
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.

Playbook
Learn how tasteful prompting helps you move beyond generic AI output by shaping context, style, and judgment from the start.

Playbook
OpenAI shipped /goal for the Codex CLI. It turns a prompt into a persisted, self-continuing contract.