Aaron Levie's on CXOTalk today, telling CIOs how to bring agentic AI into the enterprise — which finally puts someone on the hook for what happens when an agent fails. Welcome to Tech Podcast Podcast. Today: Levie on agentic governance, Mike Krieger inside an AI-native dev project, and a sharp new frame out of Europe — zombie funds. We've had three days of compute-cost math building toward a real accountability question, and Friday delivers... a Box positioning deck and a KuCoin-sponsored Instagram founder. Let's see if anybody says a number. That's the test, Joey. Levie's the first enterprise CEO this week where the frame is governance, not just capability — so does he actually answer who owns the P&L when an agent breaks? Start me with Krieger. He shipped Instagram to a billion users on basically no headcount — if anyone can tell me where the new dev cost curve actually bends, it's him. Right — the Volpi episode Wednesday danced around that number all hour. Krieger's inside Fable 5 right now, so he's the practitioner who could give us the figure. If anyone actually asks him for the figure. It's a sponsored slot — Instagram credibility wrapped around a crypto content calendar. I want the operating detail, not a badge swap. Fair worry. The line I'd actually queue first is Adebajo at Newton — she names zombie funds inside European VC specifically. Funds that can't exit, can't raise again, but haven't died. And that lands right against the Maris DPI data — small funds supposedly outperforming on distributions, while Europe's got a whole cohort that can't return a dime. Who's holding that bag? Exactly the structural problem nobody's put a number on. The privatized-gains story has a European version now — Adebajo gives it a geography. Then there's Tenderly — episode 743, simulating capital onchain. Which is, what, the fortieth base-layer-for-the-agentic-internet pitch this cycle? So press Bencic on it — what does Tenderly actually settle? “Simulating capital” has to mean something more concrete than the framing we got Wednesday. The crypto promise was cutting out the middleman, and every new layer adds one back with a license. I want to know if Tenderly's replacing intermediaries or just stacking on another one. And TBPN ran an “Is TSA Goated?” segment at 15:44 right before a Future of Airports block. Either that hides a real airport-friction mechanism or it's pure banter — and this week, hosts have been letting guests coast. If Levie names the failure mechanism today, the accountability question finally has a handle. Let's get into it. TBPN writes:
Mike Wior, co-founder and CEO of Allen Control Systems (ACS), discusses the company's development of Bullfrog, an autonomous robotic weapon station designed to counter small, fast-moving drones by combining AI, computer vision, and machine learning to achieve both speed and accuracy. He highlights Bullfrog's 100% success rate in neutralizing 13 Army-operated drones during a November field test, emphasizing its capability to address the evolving drone threat landscape.
A two-hour-twenty-four-minute show that opens with “Bezos AI Play,” detours into “Is TSA Goated” at 15:44, and then “Future of Airports” five minutes later. So is the TSA bit a real airport-tech segment or just runway to the banter? The part I'd actually queue is Mike Wior. Allen Control Systems' Bullfrog — an autonomous robotic gun against drones, and they're claiming a 100% kill rate against 13 Army-operated drones in a November field test. Right, so we go from “is taking your shoes off goated” to a robot that shot down thirteen drones with a hundred percent hit rate. That's the tonal range of TBPN in one episode. And Reger's the other one I'm watching — Neura Robotics building “Neura Gyms” worldwide to train robots on physical tasks and harvest the data. That's a concrete cost-curve detail, not a humanoid hype reel. A hundred percent on thirteen drones in a controlled test is a press-release number, though. I want the failure rate against something it didn't schedule the demo for. This one's from KuCoin:
Mike Krieger, co-founder of Instagram and head of Anthropic Labs, shared insights on Fable 5, Anthropic’s latest AI model, and its impact on software development. He noted that the fear and greed index in the tech sector is shifting as AI becomes a core collaborator, not just a tool. Fable 5 now handles complex tasks autonomously, adapting during development and executing high-level intent.
Mike Krieger co-founded Instagram, now runs Anthropic Labs — and this is landing on KuCoin's content calendar under a “fear and greed index” header. The actual episode is Dan Shipper on Every. Somebody laundered a real conversation through a crypto exchange's blog. Forget the wrapper — the title is the thing. “Lets Fable 5 code while he sleeps.” This is the practitioner we kept asking for: a guy who shipped a billion-user app at thirteen employees, now inside an AI-native dev loop. Right, but “autonomous, adapts during development, executes high-level intent” — that's a keynote slide, not a number. Did Shipper get him to say what actually breaks when it runs overnight? That's where Krieger's specific, though. The Instagram cost curve — how few people it took to scale — means if anyone can name where the new headcount math bends, it's him. I want the figure, not the vibe of the figure. And note who's framing it — head of Anthropic Labs talking up Anthropic's own Fable 5. So “AI is now a core collaborator” from the guy selling the collaborator. I'll believe the autonomy when he tells me the failure rate of the code he didn't watch get written. CXOTalk, with Michael Krigsman:
AI agents work in demos but stall in production, break existing permission models, and run up costs that CIOs never planned for in traditional IT budgets. Agentic AI in the enterprise has become a direct test for Chief Information Officers. Aaron Levie, co-founder and CEO of Box, joins CXOTalk episode 921 to discuss what agentic AI means for enterprise software, security, and the CIO.
Levie's on CXOTalk today, episode 921, and the framing is the one I've actually been waiting on all week: agents work in the demo, stall in production, break the permission model, and run up costs nobody budgeted for. All week we've had the cost-side and the fund-size data. This is the first enterprise-software CEO put in position to answer the governance question — when an agent fails in production, whose permission model just broke, and who owns that? Or it's episode 921 of a show whose whole job is making enterprise software CEOs sound visionary, underwritten by the Gartner IT Symposium. Sarah, that's a Box positioning deck with a calendar invite. The tell will be whether Levie gives a number. A failure rate, a real cost-per-agent overrun, a deployment timeline. Or does “agents stall in production” stay a vibe he sells the fix for. Here's Sifted:
Anu began her career in Sheffield in 2012, where she joined a fund and later went into the British Business Bank. After nearly five years, she moved to the LP side, where she personally deployed over £365m into funds, before moving to Atomico where she led its fund of funds strategy.
Okay, here's the one that actually woke me up today. Anu Adebajo — former Atomico partner, now running Newton — says it straight on Sifted: Europe's got a cohort of zombie funds. Funds that can't exit, can't raise the next vehicle, but haven't officially called it. They're just... shambling around the cap table. And that sits right against the Maris DPI numbers from Wednesday — if the small funds are the ones actually returning cash, what do you call the small funds that structurally can't? Adebajo's the first person this week to put a concrete structural name on it instead of vibing about “the European funding gap.” Right, and remember the line we kept hitting — gains get privatized early, risk gets socialized at the exit. The US mega-IPO version was only one form. Europe's already manufacturing the bag-holders. Brandon Zemp, writing in BlockHash:
For episode 743 of the BlockHash Podcast, host Brandon Zemp is joined by Andrej Bencic, CEO and Co-Founder of Tenderly, the simulation company for onchain institutions. An engineer by background, he co-founded Tenderly in 2018 and has spent the last eight years building it into the operational layer beneath crypto’s most sophisticated protocols, enabling engineering, finance, and risk teams to model every onchain action against the live system before any capital or customer is exposed.
Tenderly, episode 743, Andrej Bencic — “the simulation company for onchain institutions.” So you model every move against the live system before any capital touches it. Yesterday it was Kite framing itself as the base layer for the agentic internet. Today it's simulating capital onchain. Same genre, different chain — what does Tenderly actually settle? They've been at it since 2018 — eight years building the layer where risk teams dry-run a transaction before real money's exposed. That starts to sound like real plumbing. Sure, but the crypto pitch was disintermediation, right? And now the missing piece is — another layer that institutions need before they'll move. Are they replacing an intermediary or just adding one with a dashboard? The interesting tell is the timestamp at 10:46 — institutional adoption versus crypto startups. If the institutions are the ones who need the simulation layer, that tells you who's actually scared of the failure case. If Tech Podcast Podcast helps you keep up with the week, take a second to subscribe and leave a review wherever you're listening. It really helps other people find the show.
If one of today's stories is still rattling around in your head, you'll find links to all of them in the show notes. Take a look, follow the threads, and read a little deeper.
That's Tech Podcast Podcast for today. This is a Lantern Podcast.