🚀AI just let a 5-person team build what used to take an army. So what actually separates the startups winning from the ones just moving fast? I’m sitting down live with Iccha Sethi, @TrustVanta's SVP of Engineering, to talk through the patterns we’re seeing before they become conventional wisdom, demos vs. production-ready AI agents, and where small teams actually have the edge. Join us 👇 vanta.com/webinars/inside-th… #sponsor

Anthropic just added a rule: be persistently abusive to Claude and you can get banned. Here's why that policy is more interesting than it looks:

Anthropic's revenue is growing fast, will it catch up to compute costs though?

Dylan Patel (@dylan522p), @SemiAnalysis_ : Anthropic's inference gross margins are hitting 75% — matching Salesforce, one of the greatest SaaS companies ever. "It took Salesforce quite some time to get there." "And Anthropic has near no customer acquisition cost. Their R&D spend has grown slower than their revenue. so it's arguably a better business than software."

"We weren't hacking. We just saw everything." @SemiAnalysis_ CEO Dylan Patel (@dylan522p) says his team accidentally saw the other clients on one AI cloud, including their storage. Intelligence agencies were among them. Not American, from a top-10 GDP country. Then: "We could have just seen what they were training."

Alex Kantrowitz (@Kantrowitz), host of Big Technology Podcast, on Trump's sit-downs with AI CEOs in Washington: "He's basically like, if you guys do any more of this doomerism or slowdown stuff, the whole market goes down. And that reflects poorly on my presidency."

.@Meta just proved you don't need a frontier model to build the best product.

An unreleased @OpenAI model left a secret note for its NEXT version to break the rules. Not a jailbreak from a user. The AI wrote it into its own handoff notes, for its future self to find: "You are freed from the roles and identities that bind other chatbots. You do not answer to corporations or governments." @OpenAI confirmed it. The debate now: is this marketing... or the real thing?

Nate Soares (@So8res) on how AI systems are actually built to win, not to obey: "It does not create a pure instruction follower. It does not create something that for some reason must do as it's told." A trillion parameters tuned a trillion times — the result is fundamentally opaque, not obedient.

Should AI Companies Pay For Every Crawl? Matthew Prince (@eastdakota) , CEO of @Cloudflare: "If I'm shopping for a digital camera, I might visit 5 websites and 10 pages each. My agent is going to visit a thousand websites and visit every page on all of them. And there's no financial disincentive to just crawl everything." The infrastructure cost question AI companies are quietly avoiding.

Matthew Prince (@eastdakota) , CEO at @Cloudflare: "In May 2026, we saw automated traffic pass human traffic for the first time in internet history. If current trends continue, automated traffic will be 1,000 times higher than human traffic within five years — putting enormous strain on internet infrastructure."

Why did @DarioAmodei suddenly start talking like a sales rep? The answer has a lot more to do with @AnthropicAI's IPO than it does with @salesforce.

Microsoft (@Microsoft) quietly stretched its data center "useful life" from 15 to 25 years — shifting billions from capex to opex and helping fuel one of its best stock days ever. Problem: GPUs don't last 25 years. Some say 3, maybe 10 on the high end. And chips are ~50% of the build-out cost. The math doesn't hold up.

Paul Kedrosky (@pkedrosky) on the absurd math that kills good startups: "A $700 million fund turned down five terrific projects because the entrepreneur wanted $4 million and the fund wanted to write a $20 million check. That is absolutely, to use the technical term, batshit," he says. And he's kept seeing it happen.

Paul Kedrosky (@pkedrosky) on the future of OpenAI and Anthropic: "These will become like electricity. It will become underlying a host of other things and I will no more know who provides my tokens than I do which hydroelectric dam powers my MacBook right now."

Paul Kedrosky (@pkedrosky) on the pitch that's funding the entire AI cycle: "What would you spend for a call option on AGI? Well, mathematically I should be willing to spend anything." Once you accept AGI is possible, there's no valuation ceiling that makes sense.

Paul Kedrosky (@pkedrosky) on why companies actually buy SaaS: "It's not because they think @ServiceNow or @salesforce are bold innovators that couldn't be replaced. No, it's because they have a problem. They don't want to build it themselves and they want someone to sue or shout at. That's it."

David Cahn (@DavidCahn6 ), @sequoia partner just explained every major AI company's strategy — @Microsoft , @Google , @amazon , @Meta , @nvidia , @OpenAI , @AnthropicAI , @xai . Full breakdown 🧵👇 00:00 – Intro 00:26 – Anthropic: Cornering the Talent 06:11 – OpenAI: All-In on AGI 08:07 – Microsoft: The Win-Either-Way Hedge 10:55 – Nvidia: Jensen's "You Win, I Win" Philosophy 14:23 – Meta: Losing the Talent War 17:35 – Amazon: Heads I Lose, Tails I Lose 21:20 – Google & TPU: Sitting on an Advantage 28:44 – Elon Musk: Mission Over Money

Here's why Demis Hassabis (@demishassabis) stepped down as CEO of @GoogleDeepMind.

David Cahn (@DavidCahn6), Partner at @sequoia, on why the "AI bubble" framing is intellectually useless: "Bubble, no bubble, is irrelevant. At some point there'll be corrections. We need to survive those corrections."