The guy who co-wrote the Trump AI Action Plan just took a job at OpenAI — and there was exactly one window where he'd talk honestly about it. This is Tech Podcast Podcast. Today, a policy architect crosses enemy lines; then a micro-fund bets on what OpenAI won't eat, and Razorpay's CEO prices out agentic buying. Tap follow so the next episode finds you. This one comes via TBPN. So TBPN's got two items sitting right on top of each other, and nobody's saying it out loud — China weighing AI export controls, and SK Hynix eyeing a $28 billion IPO in the same week. Hynix is a memory company. HBM is the whole trade. If China moves on export controls, that $28 billion number is betting the tension goes one way, and only one way. Yeah, anchor on that number. We've spent weeks arguing about which layer captures the margin — model versus agent — and here comes a supply-chain variable that doesn't care about any of that. Banks targeting payments is the sleeper, though. Traditional finance pushing back into payment businesses while AI spend gets cut — two cost-pressure signals landing in the same week. Right, and Coinbase already trimmed AI spend 50 percent. Now banks want the payment rails back. Somebody's reading the same balance sheet as everyone else. Here's Stop the World:
The conversation covers a lot. Dean gives his views on two ideas floated by his new boss Sam Altman in the hours before recording: a global governance body for AI standards, and reports that Altman has been in talks with the Trump administration about giving the US government a stake in OpenAI.
Dean Ball was the principal author of the Trump administration's AI Action Plan, and Stop the World caught him in the window right before he started at OpenAI. That timing is the whole thing. The policy architect goes straight onto the payroll of the company his policy now governs. There's a seam there you rarely get to see. And he's reacting live to Sam Altman floating two ideas hours before tape rolled — a global AI governance body, and reports that Altman's been talking to the Trump administration about the U.S. government taking a stake in OpenAI. The stake-in-OpenAI one is wild. You wrote the rules, you join the company, and now your incoming boss wants Washington as a shareholder? I want to know what Ball actually said versus what he'll be allowed to say next month. The line that got me: he thinks government AI capability shouldn't get too far out of proportion to what everyone else can access. Coming from the guy who just wrote federal policy, that's a real position, not just a talking point. And he gives a mechanism — independent third-party auditors sitting between governments and the AI labs. That's specific. An actual check he can name, instead of vibes about safety. It ties back to the export-controls piece we just hit, too — he talks about the U.S. having incentives to withhold its best capabilities even from trusted allies, and where that leaves middle powers like Australia on data centers and rare earths. Right — the policy he wrote is now the geopolitical environment his new employer has to operate inside. He basically designed the room he's about to sit in. Here's The Ken:
And Razorpay is at the forefront of this transformation. In the last eight months alone, the company has built a UPI-OpenAI-Razorpay pilot for ordering groceries in chat, launched agentic payments across Zomato, Swiggy, and Zepto, rolled out a UPI Reserve Pay flow that practically kills the PIN, and built an agent studio on Claude for merchants to automate disputes, recoveries, and reconciliation.
Okay, this is the one I've been waiting for all week. A payments CEO actually saying what agent-native buying looks like on the ground — cake ingredients, checks your pantry, builds the cart, pays. He says that's two years out. And notice where Razorpay put itself. Underneath the transaction. The UPI-OpenAI pilot, agentic payments live across Zomato, Swiggy, Zepto — that's infrastructure choosing the plumbing layer on purpose. Which connects straight to the Shiprocket constraint — Indian SMBs wouldn't pay for software seats, only transactions. Razorpay's the layer Shiprocket runs on. So does agentic buying finally change that seat-versus-transaction math, or just pour more volume through the same pipe? I think it hardens it. If the agent is doing the searching and scrolling, there's no human staring at a dashboard to sell a seat to. You only ever monetize the transaction that clears. Right, the human's gone, so the seat's gone. And that's the tension with the TBPN piece we just hit — banks pushing back into payments the same week a payments CEO says the whole checkout page disappears. Both sides are fighting over a transaction nobody's clicking anymore. From Vanessa Larco at GTMnow:
"Investing in companies that OpenAI won't kill." That's Vanessa Larco's thesis, and in this VC bonus episode of GTMnow she breaks down exactly how she decides what survives. Vanessa, founder and GP at Premise (formerly a longtime partner at NEA, where she backed Robinhood from seed to IPO), joins Max to get into the nitty gritty of how AI is rewriting venture, distribution, and the entire go-to-market playbook.
"Investing in companies OpenAI won't kill." Finally, a thesis you can actually grade. Name the category it kills, name the one it doesn't — you can test that. It's not just a deck slide. And she's not a first-timer taking a flyer. She backed Robinhood from seed all the way to IPO at NEA — so when she leaves the big platform to run a two-person shop on a blank canvas, you don't do that as a fallback. See, that's the test I care about. A two-person fund running on memos and KPIs — either the crisp thesis keeps you alive in the squeeze, or you're the inventory everyone else is trying to clear. The line I want her pushed on is "it's just a wrapper" costing VCs real deals — she's pulling the Kayak and AWS parallels. That's the concrete part. Where exactly does the old DevTools go-to-market break, and what actually replaces it? And whether the host lets "OpenAI won't kill it" sit there as a bumper sticker or makes her draw the line on a whiteboard. If she names one company OpenAI does eat, I believe the whole framework. From Patrick O'Shaughnessy at Invest Like the Best:
Over the last 18 months, Jeremy has had hundreds of conversations with founders and with the capital behind their companies. I don't know many investors with such a high rep count in the most interesting corners of private markets, so I asked him what he has learned.
Jeremy Giffon, round two on Invest Like the Best — and the pitch is he's had hundreds of conversations with founders and the capital behind them over 18 months. High rep count in the weird corners of private markets is actually the credential I care about here. And this is the founder-side counterweight I wanted. After the Kleiner term-sheet philosophy piece Tuesday, Giffon is the guy sitting across the table from that capital — 'The Billion Dollar PDF' is his frame for how those deals actually get done. Sure, but O'Shaughnessy called him one of the most popular guests and said it's 'every bit as enjoyable.' Enjoyable is the word that makes my polished-narrative alarm twitch. I want the operating detail, not the greatest-hits reel. That's fair — but 'everyone's become subservient to the poster class' is an actual claim with edges. If he names how founder incentives bend to whoever's loudest online, that earns the hour. The poster class line I'll take. That's specific enough to be wrong, which is more than most of this week gave us. Got thoughts on today's briefing, a story we should track, or a correction we should know about? Send us a note at techpodcastpodcast at lantern podcasts dot com. We read every message.
You'll find links to every story we covered today in the show notes, so if something stuck with you, that's the place to dig in a little further.
Thanks for listening this Thursday. That's Tech Podcast Podcast for today. This is a Lantern Podcast.