Gurley: NYT profile of Dario Amodei

Bill Gurley called the New York Times piece on Anthropic's CEO "a must read for anyone associated with AI." He named corporate customers, analysts and, "most importantly," regulators of AI, because it is "critical to understand the 'why'" @bgurley. What stood out to him was Amodei as a high school student who was a "Y2K worrier," which Gurley called "quite defining" @bgurley. Reporter Ryan Mac says the team spent months on the piece. It describes a man influenced by rationalism, effective altruism and Ray Kurzweil, whose convictions "are being tested now as his company is about to go public" @RMac18. Article

Chamath: Brandur's "slightly bearish" thoughts on Stripe

Chamath Palihapitiya quoted @brandur's "overly long (slightly bearish?) thoughts on Stripe." Brandur opens by noting that roughly half of the best people he worked with at Stripe are now at Anthropic and the other half at OpenAI @brandur. Chamath's takeaway is that "the value of being public is access to capital." Public markets let a company build a warchest in exuberant times and deploy it during contractions, which private companies find much harder to do @chamath. Because a tech company is "always one turn of innovation away from irrelevance," he argues that staying private underprices its cash flows when markets are exuberant and overprices them during declines @chamath. Post

O'Shaughnessy: Ben Thompson on the AI capital gap

Patrick O'Shaughnessy posted a clip of Ben Thompson on why "we might run out of capital before we run out of demand for compute" @patrick_oshag. Thompson argues that AI spending has moved down the capital curve, from free cash flow to debt to equity. He says that if returns don't arrive in time "we could have a big blowup," and that today's spending "all goes into compute in 2029" @patrick_oshag. Post

Casado: Exo paper and poteto on formal methods

Martin Casado shared the arXiv paper on Exo, which he co-wrote with Alex Krentsel and @ankrgyl. It tests how much a model can optimize a task when it can read runtime logs and modify and rebuild its own harness. The authors borrowed from operating systems work to make those changes safely, and Casado calls the results "surprisingly strong" @martin_casado +1. Paper

He also endorsed @poteto's post on formal methods as "exactly my long standing view" @martin_casado. Poteto argues for layered checks: guardrails, empirical verification, and formal verification such as TLA+ model checking for the parts that need it. He also recommends his /statechart skill for confirming that an agent understood your intent @poteto. Post

Shorter picks

• Sarah Guo shared Alex Ker's explainer on speeding up DeepSeek Flash, saying "love us some model driven model infra" @saranormous. Ker says their internal harness optimized more than 80 kernels, reaching 991 output tokens per second. The explainer covers kernel fusion, programmatic dependent launch and host loop minimization @thealexker. Post
• Garry Tan highlighted DeepFinance as "the first long horizon benchmark for PE/IB work," adding that "solving high finance is on the path to AGI" @garrytan. It has 21 expert-built problems based on real deal data, each representing 20–40 hours of professional work @0interestrates. Leaderboard
• Elon Musk said Iain Banks "arguably has the best vision for the future" @elonmusk. He was responding to a post describing the Culture novels as a civilization of compute abundance where people choose among ships organized around different ways of life @yunta_tsai.
• Naval shared Akira the Don's new track "The Real Currency of Life," which is built on Naval's own line @akirathedon: "Not money, not time - but attention" @naval. Post
• Vinod Khosla called his WSJ video interview with Sarah Krouse on job losses, capital gains and an AI safety net a "great conversation" @vkhosla. Video
• Jason Lemkin promoted the new 20VC x SaaStr episode on OpenAI's lower $1.25 trillion valuation, Anthropic's growth "halved," and agents bypassing DoorDash @jasonlk. Episode