Today, the AI story gets out of the slide deck — reactors, security ops, and Siri actually shipping. If you're just joining, workplace AI has been sprinting past IT — Omnissa clocked AI assistant use up nearly 1000% across 2025, as employees picked tools faster than anyone could approve them. The question now is whether companies can turn that shadow-AI surge into observability and governance, rather than just slamming the door. This is Tech Podcast Podcast. Today — a security guy, a nuclear reporter, and a VC who turned down double the money, all arguing about who actually does the hard part. Let's start with the reactors. Enterprise shadow AI adoption isn't over. Follow us wherever you're listening, and the next chapter comes to you. From SC Media:
As enterprises move from AI experimentation to adoption at scale, security leaders are under pressure to enable innovation without introducing unmanaged risk. The challenge is no longer whether organizations should pursue AI, but how they can govern it, secure it, and operationalize it in ways that stand up to real-world business and threat conditions.
Okay, so ESW #468 with Keith Hollender, CEO of Arcova, treats AI security as much about execution as governance. And he's choosing that framing for a reason — it's the same diagnosis I keep hearing from infrastructure people and workspace people. Different verticals, same sentence. Right, and Hollender's worth the airtime because he's an operator — Arcova is inside enterprise environments watching teams try to connect governance, resilience, and transformation. He stays with the mechanics instead of floating up into manifesto territory. And here's the part I want someone to say out loud instead of just tease: the shadow-AI problem we've been circling all week isn't only, 'nobody's measuring the productivity.' If your people are running unsanctioned tools, that's a live security surface. The execution gap is the attack surface. Which lands harder when you remember how many enterprise pilots die before production. A lot of governance frameworks treat the problem as policy. Hollender's saying the breakdown happens in the doing — and that's why the pilots stall out. From Alex Roy at Autonocast:
Alex, Ed, and Kirsten sit down with George Kalligeros, CEO and co-founder of Aseon Labs, to explore the overlooked infrastructure required to scale autonomous vehicles and robotaxis. The discussion covers robotic vehicle servicing, decentralized charging, autonomous fleet maintenance,…
Autonocast #368 — Alex Roy talking to George Kalligeros, co-founder of Aseon Labs, and the whole pitch is AV operations. They're putting the spotlight on the fleet grind around the perception stack and the model. Which is the part everybody skips. The driving demo goes viral, and then nobody wants to talk about what it takes to actually keep a fleet running once the cameras are off. And it rhymes with the ESW segment we just hit — Hollender kept pulling AI security back to execution, and Kalligeros is doing that for AVs. Two totally different verticals, same morning, same pain point. Right — I'd want Roy to press Kalligeros for a real operating number. How many remote operators per vehicle? What's the intervention rate? That's the stuff that decides whether the unit economics ever close. From Geoff Brumfiel at NPR:
A race started by President Trump is leading to the rapid construction of new, experimental nuclear reactors. If they work, they could power data centers for artificial intelligence. But critics worry that the breakneck pace is compromising safety and public trust.
Okay, this is the one I flagged coming in. NPR's Short Wave — Geoff Brumfiel — on experimental reactors getting rushed online to feed AI data centers. Valar Atomics already has one running. And this is the physical underside of the whole week. We've been telling the AI-stack story through economics and enterprise pilots — and we haven't once touched the power grid it runs on. Right, and the part that sticks with me is the critics saying the breakneck pace is compromising safety and public trust. All week, AI risk has meant legal risk or financial risk. This is the first story that puts physical risk on the build-out. Brumfiel gives it weight because he isn't just running a booster segment. He keeps the tension in the room — build fast for the data centers, or protect the trust it takes decades to earn back. That's the tradeoff, stated plainly. From Federico Viticci and John Voorhees at MacStories:
This week, on AppStories, Federico and John look at what Siri AI is good at, where it needs improvement, and where it might be heading. On AppStories+, the conversation expands to Federico and John’s experiences with the betas of macOS Golden Gate, iOS 27, and iPadOS 27.
AppStories 494 — Federico and John doing the full Siri AI teardown: what it's actually good at, and where it falls down. After a week of treating AI as an execution problem, here's the execution problem shipped straight to your pocket. And it's framed around Apple's trickle-up strategy — the assistant gets smarter on device first, then the rest of the OS inherits it. The claim to test is that system-wide handoff, beyond whatever looked good in the demo reel. The token-efficiency crowd's been arguing all week that smaller and leaner beats token-maxing. Siri's the consumer test case — does any of that theory cash out when your phone has to answer 'set a timer' without embarrassing itself? What I want from those two is the specific where-it-breaks list. Apple's line is 'profoundly more capable and personal.' Federico's usually the one who names exactly which task still fails. Right — the value here is 48 minutes of two people who actually live in the betas telling you which promises survived contact. That beats another keynote adjective. Rohin Dharmakumar, writing in The Ken:
He rejects the venture power law: rather than chase one 100X outlier, Fireside works hands-on through a "centre of excellence" so that a high share of its companies succeed, about 60% in the first fund rising to 80%-plus later
Finally. After a week of decks carrying the story and valuations floating on vibes — here's a guy who turned down twice the money. On purpose. Capped fund four at the size of fund three. And the reasoning is concrete, which is why it lands. Singh says a bigger fund breaks the early-stage model that actually produces his returns. He sounds like an operator protecting the model, instead of flexing a fundraise. He literally calls it an anti-power-law fund. The whole 20VC-SaaStr gospel is 'find your one 100X and eat the losses.' Singh's running the inverse — 60% of the first fund succeeding, up past 80% later. I keep coming back to 2017. He couldn't sell global institutions on an Indian consumer story, so he raised from the families who'd actually built the brands — the Mariwalas, Premji Invest, Unilever, ITC. That cap table tells you the thesis. Those are the receipts. He took money from people who'd already won the consumer bet instead of just pitching it to them. Hard to argue with the reference check. Got thoughts on today’s briefing, a tech story we should track, or a correction? Send us a note at techpodcastpodcast at lantern podcasts dot com. We read every message, and your tips help shape the show.
You’ll find links to everything we covered today in the show notes, so if one of these stories grabbed you, that’s the place to dig in a little deeper.
That’s Tech Podcast Podcast for today. This is a Lantern Podcast.