Sovereign AI: America Built an AI Kill Switch — Then Aimed It at Allies

Transcript

Sam: A week or two ago, the United States quietly switched off its most powerful AI for an entire continent of its closest allies — and the thing that got switched off wasn't a chip, or a server, or a building.

Alex: It was permission. One directive from Washington, and a model that millions of people were using vanished for everyone outside the US — friends included — inside twenty-four hours.

Sam: So America built a kill switch for AI. And then flipped it on its own allies.

Alex: Welcome back to Dan's AI Intel — the show where we try to make sense of the fastest, most consequential shift any of us are ever going to live through. The stuff that's actually reordering power while everyone's arguing about chatbots.

Sam: I'm Sam, here with Alex, and today we're getting into the day a piece of commercial software got treated like a guided missile.

Alex: Here's the shape of it. A week or two ago, the US ordered its leading AI lab to cut its most capable models off from every foreign national on Earth. We'll get into what actually happened — the mechanism is genuinely wild — and what the order really achieved versus what it claimed.

Sam: But the reason this one is the one I haven't been able to stop turning over isn't the news itself. It's the question it cracks open underneath. Because if access can be switched off from another capital, with a phone call, then what does "sovereignty" over AI actually mean anymore? Like — what does owning any of this even buy you?

Alex: And that question takes us on a tour. We'll run that lens over Europe, over India, over basically everyone who isn't the US or China. We'll look at who's quietly holding leverage nobody's talking about —

Sam: — and we'll get to the one move almost every country on Earth is making right now that, when you actually look at it, is exactly backwards. There's a ladder everyone's climbing, and I came away pretty convinced it's the wrong ladder. But I'm not going to spoil where that lands.

Alex: If you've been enjoying the show, do follow us on Spotify or Apple Podcasts so you don't miss the next one — there's a new one every week and they build on each other.

Sam: Okay. Let's start with the directive itself, because the design of it is the whole story. Walk me through what actually happened.

Alex: So for about thirty years, American export controls on advanced technology have pointed at physical things. Machine tools. Strong encryption. And since 2022, the high-end chips that train large AI models. The logic was always the same and always tangible — control the object, slow your adversary down.

Sam: Right. You can't ship the thing, the other guy can't build the thing.

Alex: Exactly. On June twelfth, twenty twenty-six, a Commerce Department directive completely inverted that logic, in a way that really has no precedent. The object of control wasn't the chip. It was the capability the chip produces — a finished, deployed model's API. The thing you reach from a browser, anywhere with an internet connection.

Sam: Hang on. So not the factory, not the machine — the thing the machine makes, after it's already made. How do you even export-control something that's just... sitting on a server answering questions?

Alex: That's the genius and the menace of it. And the mechanism is worth slowing right down on, because the design is the story. It was issued under the Export Administration Regulations, run by the Bureau of Industry and Security. Though here's the first odd note — a law firm tracking it pointed out that, quote, "the specific statutory basis has not been publicly identified with precision." No formal notice. No published technical review. No comment period.

Sam: So even the lawyers couldn't fully tell you which law it was standing on.

Alex: Couldn't pin it down. But what makes it radical is that it reaches into a doctrine called "deemed exports." This is an old rule — the idea that if you give a foreign national access to controlled US technology while they're standing inside the United States, that legally counts as exporting it to their home country.

Sam: Okay, so think of it like... the technology has a passport check built into it. It doesn't matter where you physically are — it matters where you're from.

Alex: That's exactly it. For decades that doctrine governed who could stand next to a centrifuge, or read a satellite blueprint. Genuine weapons stuff. Now apply that same rule to a chatbot.

Sam: Oh no.

Alex: And you get the absurd-but-completely-logical conclusion that an engineer on a work visa, sitting at her desk in San Francisco, at the very lab that builds the model — she cannot be shown the model she helped build. Because letting her type into it is, on paper, an export to her passport.

Sam: That is genuinely insane. She built it and she's not allowed to use it because of where she was born. So what does a company even do with an order like that? You can't exactly card everyone mid-conversation.

Alex: You can't. No provider can reliably sort foreign nationals from American ones inside its own live user base, in real time, at scale. So the only move that actually complies is the blunt one. Within a day, Anthropic disabled its two most capable models — Fable 5 and Mythos 5 — for every customer on Earth. Allied, domestic, everyone. Because it had no other way to obey.

Sam: So to comply with "don't let foreigners use it," the only button that actually exists is "nobody uses it."

Alex: Nobody uses it. And here's the line I keep coming back to. The kill switch that policy people had theorised about for years — they imagined some elaborate technical apparatus, some deep architectural thing. It turned out to be a single directive from Washington and a company with no choice. That's the whole machine.

Sam: And once you've watched it work once...

Alex: You can never again pretend it isn't there. The precedent is the product. This wasn't the last order like this — it was the template for the next one.

Sam: Okay so that's how the switch got flipped. But let me push on the why, because they didn't do this for fun. What was the actual justification — and did it hold up? Because you're hinting it didn't.

Alex: So the stated reason was national security. There were reports that a China-linked group had exploited the models. And the response from Anthropic is where the real significance hides — because they didn't just object to the decision. They undercut the entire premise of it.

Sam: How so?

Alex: They said they'd received, quote, "only verbal evidence of a potential narrow, non-universal jailbreak." And then, more damaging — they said the capability in question, quote, "is widely available from other models, including OpenAI's GPT-5.5, and is used every day by the defenders who keep systems safe."

Sam: Wait. So the company being told "your model is too dangerous" turned around and said — first, you only told us about it out loud, you never showed us proof, and second, the scary thing it does, our competitor's model does too, and it's a normal security tool?

Alex: That's exactly what they said. And sit with that for a second, because it dismantles the headline. The model that triggered the single most aggressive software export control in history was, by its own maker's account, not dramatically ahead of what competitors already ship to the same customers.

Sam: So if the dangerous thing is also in GPT-5.5, and it's in the open models anyone can download —

Alex: — then banning one API doesn't meaningfully deny that capability to anyone who actually wants it. A determined adversary just walks next door. Which means the control was never really about denial.

Sam: So if it doesn't deny anything... what's it for? Because nobody fires the biggest gun in the arsenal to accomplish nothing.

Alex: It was about demonstration. It establishes — out loud, in front of everyone — that Washington can reach into a private commercial product and switch it off. And is willing to, even when the model is no special weapon, and even when the people on the wrong side of the line are friends.

Sam: It's a show of force. The point isn't the target, the point is that everyone's watching.

Alex: And there's a second-order thing here that I think most of the coverage just skated past. The order quietly redefined what an AI lab even is. A frontier model used to be a product its maker controlled. After June twelfth, a frontier model is an asset a government can pick up and use as an instrument of statecraft — with the lab as the reluctant operator of the switch.

Sam: Oof. So Anthropic's loud public fury wasn't really corporate theatre. It was them realising —

Alex: — that their core product had just been nationalised. In function, if not in name. And that every customer relationship they have, especially every foreign one, now carries a sovereign-risk asterisk they can't remove. That's the part that travels around the world. Every government watching now understands the AI it depends on lives inside someone else's regulatory perimeter.

Sam: Okay, that's the thing that stuck with me from that whole stretch — "nationalised in function if not in name." That's the line. So we've established what the switch is, and that the security story doesn't really hold. Which makes me want to follow the money. If it's not about denial — what did Washington actually buy here, and who's paying for it?

Alex: Great instinct, because the order behaves like an economic move way more than a security one. Here's the backdrop. The US has spent two years trying to deny China frontier capability through hardware controls — the chip embargo. And the chip embargo underperformed.

Sam: How do you mean, underperformed? I thought that was the big lever.

Alex: China smuggled, China stockpiled, and crucially China stood up Huawei's domestic Ascend chip line. Still only around sixty percent of an H100's inference throughput — so, behind — but improving, and, this is the key word, uncontrollable from Washington. You can't embargo a chip they make themselves.

Sam: So when your embargo on the thing starts leaking...

Alex: ...the strategist's next move is to control the layer above the thing. You couldn't stop the chips, so you go up a floor and grab the model. June twelfth is that move.

Sam: But hold on — wouldn't a model ban leak even worse than a chip ban? A chip is a physical object you can at least try to stop at a border. A model is just... a file.

Alex: You've gone straight to the punchline. A model ban is way leakier. You cannot un-train a model. The weights diffuse, the open Chinese models are genuinely competitive, and — by Anthropic's own testimony — the capability was already available elsewhere. So this order will not reliably keep frontier capability away from a state-level adversary. China routes around it by lunchtime.

Sam: So it fails at the thing it says it's for. Then I'm back to my question — what does it actually succeed at?

Alex: It demonstrates leverage over friends. Look at who got cut off inside twenty-four hours. It wasn't Chinese labs — they dodge controls as a matter of routine. It was allied governments. Allied companies. Visa-holding researchers physically inside the United States. Australia — an AUKUS partner, deep in joint AI defence work with the US — was reportedly locked out with, quote, "no warning and no backup plan."

Sam: Their own defence partner. No heads-up.

Alex: None. So the functional achievement of the order is a live, witnessed proof that America holds a dial on the rest of the democratic world's access to the best AI — and will turn it. That's a leverage asset. And leverage is the single most valuable thing to hold in any negotiation.

Sam: Okay so the US executive branch pockets this leverage. Who eats the cost?

Alex: And notice one thing about that leverage first — it's contractual and revocable, not structural. There's no statute being tested in court, no settled rule. There's an order one administration issued, that a future one could expand, or narrow, or weaponise completely differently. Hold onto that, it matters later.

Sam: Noted. The leverage itself is kind of... improvised.

Alex: Exactly. Now, the cost. It lands on three groups. First, the allied firms and agencies that built their whole workflow on a model they can be barred from overnight — that's the Australia situation. Second, it lands on Anthropic, whose foreign enterprise business now carries a political-risk discount that its American-only rivals just... don't have.

Sam: Right, why would a foreign government sign a big contract with the lab that just demonstrated it can be ordered to cut them off.

Alex: Precisely. And the third one is the subtlest and I think the most important. It lands on the idea of the contract itself. Watch the natural allied response — the G7 immediately scrambled to assemble a "trusted partner" list. A roster of nations who'd be allowed to buy back access.

Sam: Oh, that sounds like the fix though. Get on the approved list, you're safe again.

Alex: It looks like the remedy. It's actually the bill. Think about what a vetted roster of countries permitted to purchase frontier access really is — it's the formalisation of dependency, not the cure for it.

Sam: So... think of it like a nightclub that just threw you out, and the "solution" is you can reapply to be on the guest list — that they control, and can revise whenever they want.

Alex: That's the exact dynamic. Access you have to re-apply for at the start of every administration is not an asset you own. It's a subscription. With a sovereign counterparty who can change the terms — and you just found out the terms can change with a single phone call.

Sam: That reframe is brutal. The cure is just the disease with paperwork. Okay — so let's bring it to ground. We keep saying "allies," "Europe." Where does Europe actually stand when the music stops? Are they exposed, or do they have cards?

Alex: Honestly? Both. Exposed, but not powerless — and the entire European story lives in the gap between those two words. Let me start with the one number that frames everything else. Europe controls roughly five percent of global AI compute. The United States controls around eighty.

Sam: Five versus eighty. That's not a gap, that's a different sport.

Alex: It is. And the EU's flagship answer — the AI Gigafactories initiative, inside a bigger program called InvestAI — puts up on the order of twenty billion euros of public money. Which sounds enormous, until you set it next to one single year of American hyperscaler spending.

Sam: Go on, hit me with it.

Alex: Amazon alone has signalled around two hundred billion dollars of capital expenditure for twenty twenty-six. Microsoft near eighty billion. Alphabet around seventy-five. Meta in the sixty to sixty-five billion range.

Sam: Wait — so one American company's spending for one year is something like ten times Europe's entire flagship, multi-year program?

Alex: Roughly ten times. And several of the European gigafactory projects are already slipping their timelines. The arithmetic is just brutal and it does not bend. You cannot buy your way to compute parity with twenty billion euros, and no amount of strategic urgency changes the multiplication.

Sam: Okay so here's where my brain goes, and I want you to argue the other side properly. Surely the answer to "we got switched off" is "so build your own, then nobody can switch you off." That's the obvious move. Why isn't that just... right?

Alex: And you should push there, because the build-it-yourself camp is not foolish — it deserves a real hearing. The serious version, made by analysts in Australia and elsewhere who literally watched June twelfth happen to them, goes like this: dependency is the disease, and the only durable cure is owning the whole stack. Your data, your compute, your model, your policy, in one coherent national ladder.

Sam: And that's got real force, right? If you own none of it, you've got no fallback when the switch flips.

Alex: None. And "just use the free open models" is cold comfort if you don't have the compute to run them at scale, or the talent to keep them alive. The build case is genuinely strongest exactly where you'd expect — the workloads that cannot tolerate a foreign perimeter at all. Defence. Intelligence. Critical infrastructure. There, "good enough and fully under my control" beats "best in class but revocable" every single time. Any honest take has to grant that domestic capacity is a real instrument of sovereignty.

Sam: Okay, good, so it's not nothing. But you're setting it up like there's a "but" coming.

Alex: There's a but coming, and it's the most interesting strategic call in all of European AI. Watch what the smartest European player actually did with that argument. So — Mistral. Mistral was supposed to be Europe's frontier answer. And for a while it was gloriously loud. Over four billion dollars raised. Around a fourteen billion dollar valuation, with NVIDIA anchoring the round. An eight hundred and thirty million euro debt raise for thirteen thousand eight hundred top-end GPUs near Paris. A sovereign cloud. Data centres in Sweden. Real infrastructure, real ambition.

Sam: That sounds like exactly the build-your-own dream. So what happened?

Alex: The frontier noise went quiet. And the reason is the fascinating part. Mistral pivoted. From "beat GPT" to — and this is the new pitch — "be the sovereign, full-stack platform that Europe's militaries, and utilities, and banks can run on their own premises."

Sam: Hm. On first hearing that sounds like giving up. Like, they couldn't win the race so they changed the subject.

Alex: That's the trap reading, and it's wrong. It's not failure dressed up as strategy. It's a company realising that a marginally better frontier model is not a durable European advantage — because the US can always out-spend the next training run. They'll just buy the next one. But being the provider Washington physically cannot switch off — that's an advantage that compounds. And June twelfth just made it priceless.

Sam: Ohh. So Mistral stopped competing on the one axis where it could only ever lose —

Alex: — and started selling the one thing the export order had just made scarce. Un-switchable-ness. They read the room better than entire governments did.

Sam: That's such a sharp move. Okay, what about the big national programs — France, Germany, the UK. Are those the real thing, or more monuments?

Alex: They're serious commitments by serious states — and they all carry the same asterisk, which is the tell. France headlines a blended hundred-and-nine billion euro AI investment figure, and quadrupled its Jean Zay supercomputer to around a hundred and twenty-six petaflops. Germany runs the JUPITER exascale machine and is standing up an "Industrial AI Cloud" with Deutsche Telekom and NVIDIA. The UK's put over one point one billion pounds into sovereign AI infrastructure, plus a roughly seven hundred and fifty million pound national supercomputer.

Sam: Those are big real numbers. So where's the asterisk?

Alex: Nearly every one of them is built on NVIDIA chips. Frequently running on US hyperscaler clouds. The exact dependency June twelfth exposed. So picture it — a sovereign supercomputer, on European soil, running rented American silicon, to serve a model your own country might be legally barred from using.

Sam: So it's sovereignty with an asterisk. The building's in Frankfurt but the control still lives somewhere else.

Alex: The hardware's on European soil. The control terminates in Washington, in Santa Clara, and in Hsinchu, Taiwan. The flag on the roof is European. The dials are not.

Sam: Okay. That word "asterisk" keeps coming up, and it's making me suspicious of the whole project. Because if every one of these expensive national builds still has the asterisk... is everyone just doing the wrong thing? Like, at scale?

Alex: And that is exactly the right suspicion, and it's the heart of all of this. There's a reflexive sovereignty ladder that gets recited everywhere from Canberra to Brussels. It goes: data sovereignty, then compute sovereignty, then model sovereignty, then policy sovereignty. Climb the rungs in order, arrive at independence.

Sam: That sounds completely sensible, honestly. What's wrong with it?

Alex: For most countries it's the wrong ladder. And June twelfth is the clean proof, because it lets us run the experiment. Ask one precise question. Would a data centre on your own soil have actually helped you on June twelfth?

Sam: Huh. Let me think. The thing that got switched off was... permission to use a specific model. Not the hardware.

Alex: Right. So a German firm renting top-end chips in Frankfurt was exactly as cut off from Fable 5 as one renting them in Virginia. Because the control attached to the model, not the location of the silicon. Owning the building would not have returned your permission.

Sam: So the thing everyone's spending billions on... wouldn't have solved the actual problem that just happened to them. That's wild.

Alex: It gets worse, because the building isn't neutral — it's dependency rendered in concrete. Think about what a "sovereign" data centre actually is. It's a stack of NVIDIA chips, fabricated on TSMC's lines, in Taiwan, using ASML's lithography machines, from the Netherlands, very often orchestrated by a US hyperscaler's software. A supply chain that terminates in two or three foreign chokepoints, no matter whose flag is on the roof.

Sam: So "compute on my soil" feels like control —

Alex: — and is mostly theatre. Because every layer beneath the floor tiles is owned somewhere else. The sharpest line I've seen on this puts it flatly: a gap in owned compute, quote, "does not automatically translate into a gap in usable compute access." What actually matters is "access, contractual control, upstream bottlenecks, power costs and alliance."

Sam: Okay so if pouring concrete is the wrong move — and I'm dying to know what the right one is — what is the actual edge? Does Europe have anything real here, or is it just exposed?

Alex: This is my favourite turn in the whole thing, because the real European edge is the exact opposite of a compute empire it can't afford. It's indispensability. The chokepoints Europe already holds over everyone else.

Sam: Wait, Europe holds chokepoints? I thought they were the ones getting squeezed.

Alex: They hold huge ones, they just don't think of them as weapons. ASML — Dutch company — has an effective monopoly on EUV lithography. Which means no advanced chip, anywhere on Earth, gets made without a European machine. None. IMEC, in Belgium, sits upstream of the entire industry's research. ARM — British — its chip architecture is in nearly every device on the planet.

Sam: So these aren't workloads Europe runs. These are... dials Europe holds over everybody else.

Alex: That's exactly the reframe. They're not things Europe uses. They're things the world cannot proceed without. And that completely changes what the real goal should be. It's not "a data centre in my country." The real goal is guaranteed, secure, private, un-revokable access to frontier capability — with real fallbacks — plus enough leverage of your own that cutting you off costs the other side something too.

Sam: So you don't have to be self-sufficient. You have to be too painful to switch off.

Alex: Too painful to switch off. And notice — domestic compute is one instrument toward that. Valuable for the sensitive workloads that genuinely need it. But it's not the destination. And this is how you finally settle the argument we set up earlier. The build-domestic camp is right that dependency is dangerous and some capacity must be owned. But they draw the conclusion one rung too high on the ladder. They conclude "therefore build the whole stack." When the evidence actually says — "therefore secure un-revokable access, hold leverage, and build only the slice you genuinely cannot otherwise control."

Sam: So it was never really ownership versus access.

Alex: That's the deepest part. It was never ownership versus access. It's that access plus leverage is the achievable form of the thing ownership was only ever a proxy for in the first place — which is control over the outcome. And the G7 trusted-partner list is the giveaway. It's the build-domestic instinct curdling into managed dependency. A roster you petition to join, instead of a position you hold. Sovereignty you have to re-apply for is just the absence of the thing, with a nicer name.

Sam: Okay. So everyone's climbing the wrong ladder. That's a great diagnosis — but it'd be a cop-out to stop there. What should countries actually do? And I assume it's not one-size-fits-all.

Alex: It is not, and that's the whole point — the right move depends entirely on the hand you're holding. There's a clear split, and it's by tier. Start with the big European states — Germany, France, the UK. The advice is: stop chasing a frontier parity you cannot fund. Chase indispensability and insurance instead.

Sam: Which means what, concretely? Because "be indispensable" is easy to say.

Alex: Three concrete things. One — pool, ruthlessly. A genuine pan-European compute pool beats any single national budget, and fragmentation is the one self-inflicted wound Europe can actually fix tomorrow if it wanted to. Two — back the strategy Mistral already chose. The sovereign, on-premises, un-switchable stack that ministries and enterprises can run inside their own walls. Instead of financing a doomed frontier relaunch for national pride. And three — weaponise the chokepoints you already own. Make ASML and ARM a two-way threat, so that "we can cut you off" is a sentence Washington also has to weigh.

Sam: So the goal isn't to be self-sufficient. It's to be too entangled to embargo.

Alex: Too entangled to embargo. Which, conveniently, is also far cheaper and far more achievable than self-sufficient. Now — India. India holds a much stronger hand than its raw compute share suggests.

Sam: Why? On paper they're not a compute superpower.

Alex: Because they've got three things money can't quickly buy — scale, talent, and a deliberate sovereign play already running. IndiaAI offers something like thirty-four to thirty-eight thousand shared GPUs at subsidised rates. Indigenous foundation models. There's Sarvam, there's a multilingual model called Param2. The smart move for India is not to try to out-train Fable 5 — that's a losing race.

Sam: So what do they do instead?

Alex: They own a good-enough, fully-controlled, open stack — in India's own languages, on India's own compute. And then they use the talent pipeline and the sheer size of the market as the leverage that keeps frontier access flowing on India's terms, not on a vendor's. Because — and this is the nice twist — scale is a chokepoint too. When your market is big enough, no frontier lab can actually afford to be locked out of it.

Sam: Oh, I like that. Being a giant customer is its own kind of power. Okay, last tier — what about the small and middle countries? They can't out-spend anyone. Are they just... stuck depending on whoever?

Alex: And here's the genuinely hopeful part — they cannot reach self-sufficiency, and crucially, they don't need to. The achievable goal for them is engineered un-revokability. Three moves again. First — lean on open-weight models you host yourself. Switzerland's model, Apertus, is the template. They released both the weights and the training data, fully open.

Sam: And why does physically holding the weights matter so much?

Alex: Because once the weights are sitting on your own machines, there is no foreign regulator who can switch them off. There's no API to revoke. June twelfth literally cannot happen to you for that model. That's the whole game.

Sam: That's the un-switchable thing again. You're not borrowing the capability, you're holding it.

Alex: You're holding it. Second move — pool compute and model development across like-minded states, so the unit of sovereignty is a bloc, not one small national budget. And third — negotiate access collectively, because the indispensability of a coalition beats the supplication of any single member standing alone at the door.

Sam: So even for the little guys, the takeaway isn't "go build a data centre."

Alex: It's never "build a data centre." For this tier it's "make sure the capability you depend on is one that nobody else holds the only key to." And across all three tiers — big Europe, India, the small states — the principle is identical, only the instruments change. Secure the access you cannot live without. Hold leverage of your own. And build only the slice you genuinely must control. The countries that mistake the building for the outcome will spend the most, and end up the most exposed.

Sam: Okay. Let me try to pull the whole thread together, because we've covered a lot of ground. Tell me if I've got it.

Alex: Go for it.

Sam: So — June twelfth, the US cuts its most powerful AI off from the whole world, allies included, in a day. And the thing they actually switched off wasn't hardware. It was permission. The security reason for it didn't really hold up — the maker said the scary capability is in competitors' models too — so the order's real product wasn't denial, it was a public demonstration that the switch exists and that being a friend doesn't make you exempt.

Alex: That's it exactly.

Sam: Then the instinct everyone has — "fine, we'll build our own" — turns out to be mostly the wrong ladder, because a data centre on your soil running borrowed chips and a switchable model is just dependency with a flag on it. And the real, achievable goal isn't owning everything. It's un-revokable access plus leverage — open weights you actually hold, chokepoints like ASML you can push back with, coalitions too entangled to embargo. Mistral already figured this out and pivoted to exactly that.

Alex: You've got the whole thing. And if I had to leave a listener with the two or three things to actually walk away holding — first, sovereignty over AI is not about whose soil the compute sits on. It's about whether anyone else holds a dial on your access. Second, owning the building is the proxy people reach for; access-plus-leverage is the thing the proxy was standing in for all along. And third — the one fact that would flip this whole conclusion is if a frontier capability gap got so wide that a US model became genuinely irreplaceable. But on the current evidence — competitors at parity, open weights catching up, capability diffusing everywhere — that gap is narrowing. Which is exactly why the smart bet is leverage, not concrete.

Sam: And honestly — that's the thing I'll be carrying around. This was the moment AI stopped being a product and became a sanctioned good. The first frontier model export-controlled like a weapon, aimed first at America's closest friends.

Alex: And that's it for today. Thank you so much for listening. I really hope you came away seeing a bit more clearly where all of this is heading — because it's a genuinely complex, fast-moving picture, with a brutally short shelf-life on what any of us knows. That's exactly what makes it worth following this closely. If you've enjoyed it, please do consider giving us a follow.

Sam: And a quick note, for full transparency: this show is AI-generated. Dan builds a custom stack of AI tools to chase the questions he can't stop thinking about — it actually started out made with NotebookLM, and it's now produced with his own engine — mainly so he can learn this stuff himself, and he publishes it for anyone who'd like to come along for the ride.

Alex: And one more, from Dan directly: he'd genuinely love to hear from you. What you'd want more of, what you'd change, what landed and what didn't. You can write to him at podcast@connectiveshift.com. We're really looking forward to hearing from you.

Sam: Thanks for listening. We'll see you next week.