Ep 12 – Own Your Stack: What the Claude Fable 5 Shutdown Should Teach Every Founder

Claude Fable 5 has been dark for over a week. Amazon an $8B investor in Anthropic flagged a guardrail breach to the US government, and the model got pulled from export entirely. 48 hours later, China shipped a frontier model that topped the benchmarks at a fraction of the cost.

Harry and Anthony break down the “Sad Hill standoff” between Anthropic and Washington, why Amazon went to the government instead of Anthropic directly, and what it means that export controls have now reached model usage — not just code. Then: the case for owning your own prompt and stack instead of renting frontier intelligence, why AI ROI still isn’t showing up on balance sheets, and a real-world look at where AI actually pays off (and where it doesn’t, looking at you, Zillow).

00:00 – Claude Fable 5 still offline, one week later

02:00 – Karpathy’s reaction, China ships a rival model in 48 hours

03:00 – The Amazon jailbreak report and the Anthropic–DC standoff

16:00 – Regulatory capture, Dario’s public benefit corp bet

30:00 – Why “own your own stack” matters more than ever

34:00 – The real AI ROI question — who’s actually making money

59:00 – Zillow’s AI misstep and the lesson for operators 1:01:00 – Pliny the Elder, and a wrap for the week