{"slug":"building-defense-technologies-protect-democracie","show_id":"davidsenra","title":"Building Defense Technologies to Protect Democracies | Torsten Reil, Helsing","display_title":null,"title_orig":"David Senra × Torsten Reil","dek":"Helsing argues GPS-free, AI-enabled strike drones are making tanks and artillery obsolete, with 100,000 systems potentially deterring a European land war. HX-2 output has scaled from dozens monthly to a 1,000 baseline and 3,000 capacity, while Centaur took full control of a Saab Gripen after beating pilots in simulation. Germany’s defense budget may exceed €180B, but procurement speed and unresolved machine-life-and-death ethics remain key variables.","lang":"EN","date":"2026-08-26","published_at":"2026-08-26T12:00:35+00:00","duration_min":84,"source_url":"https://www.youtube.com/watch?v=OduvOCrVpgo","source_label":"YouTube","rss_url":null,"youtube_id":"OduvOCrVpgo","youtube_url":"https://www.youtube.com/watch?v=OduvOCrVpgo","thumbnail_url":"https://i4.ytimg.com/vi/OduvOCrVpgo/hqdefault.jpg","chips":["person:David Senra","person:Torsten Reil"],"provenance":[["SOURCE","YouTube captions"],["SYNTH","claude-fable-5"],["CLEAN","gpt-5.6-luna"],["VERIFY","gpt-5.6-luna"],["PARTICIPANTS","gpt-5.6-luna"],["TRANSLATE","gpt-5.6-luna"]],"title_alt":"打造守护民主的防务技术｜Torsten Reil，Helsing","display_title_alt":null,"dek_alt":"无需GPS的AI攻击无人机或将终结坦克时代，10万架足以威慑欧洲陆战。HX-2月产能已达1000架基准、可扩至3000架，Centaur曾完全接管Saab Gripen。德国防务预算或超1800亿欧元，但采购速度与机器生死决策的伦理仍是风险。","lang_alt":"ZH","tldr_md":"- **Reil's headline call: the era of tanks and artillery is over, and 100,000 GPS-free, target-finding strike drones at a border would deter a land war in Europe.** A Russian soldier's survival at the front is apparently \"about 20 to 30 minutes,\" though Reil calls that a ballpark; 3,000 invading tanks facing 100,000 drones \"will destroy those tanks, I think, in a matter of hours.\" He caught flak for the claim a year ago — \"I still think that's the case.\"\n- **Jamming, the standard objection to drone dominance, is increasingly circumvented.** FPVs can use fiber-optic cables while larger drones navigate GPS-free with AI that works \"in daylight, at night, in the snow,\" even over shell-deformed terrain — so warfare becomes \"machines versus machines,\" a battle of materiel where sending humans is both suicidal and, in Reil's word, \"unethical.\"\n- **Helsing's AI pilot beat pilots in simulation and then took full control in a real fighter-jet test, weakening the case for crewed jets.** Trained with AlphaGo-style reinforcement learning in a simulator running 20–25,000x realtime, it took full control of a Saab Gripen over the Baltic Sea last year and flew three air-combat scenarios against a human-controlled Gripen — the \"last bastion\" of human intuition \"turns out to be [bull].\" Helsing then bought an aircraft manufacturer outside Munich; the CA-1 Europa flies next year and is designed to launch from motorways because airbases will be attack targets.\n- **The manufacturing thesis: hardware \"good enough,\" complexity absorbed into software, then pure mass.** HX-2 strike-drone output went from dozens a month to a 1,000 baseline and can now scale to 3,000; the bottleneck is contracts, not factories — a €1B German framework just closed. Reil said Germany's defense budget was €52B last year, originally projected at €153B in 2029 and perhaps now above €180B; with no more soldiers most likely, most of the money would go to equipment.\n- **Helsing was founded in March 2021 when banks bucketed defense with \"pornography and gambling and smoking\" — some people close to the company had been debanked and no VC would engage, so the €8M seed came from individuals and family offices, then Daniel Ek put in €100M of his own money (\"ideally 100 million euros,\" a number Reil eyeballed).** That war chest funded the program-of-record strategy: Helsing won the Eurofighter electronic-warfare upgrade with Saab as the dark horse 2 years 4 months after founding, which Reil thinks made it the fastest company to reach a program of record.\n- **The exit stance is categorical: \"we will never sell the company.\"** A few months in, the founders chose to \"go full throttle and we either completely crash the thing or be super successful\" — and Reil argues Europe's real deficit is not regulation or tax but founder aggression: it's \"95% of a company's success... everything else is excuses.\"\n- **Operationally: talent density above everything — Helsing's first contract, signed before incorporation, was with a recruitment agency.** Constant pipeline pressure makes firing a \"low opportunity cost thing,\" quarterly calibration reviews hold managers accountable for their ratings, and Reil hires senior people on 10-minute reads for \"effortlessness\" — deep knowledge plus curiosity plus high agency. He recalls no such miss at Helsing, but three at NaturalMotion.\n- **The best outcome for Helsing is that nothing happens: deterrence via \"the cost calculus on the other side.\"** Reil's unease is that journalists grilled him on killer robots and red lines before the Ukraine war and \"no one asks these questions anymore,\" leaving machine life-and-death ethics to the company's own \"inner discipline.\"\n","digest_md":"### 1. The era of tanks is over — front-line survival is measured in minutes\n- Reil's opener: in Ukraine \"tanks are barely being used anymore,\" and the same holds for artillery. A Russian soldier's survival time at the front is apparently \"about 20 to 30 minutes\" — a ballpark, he concedes — and worse in a tank, a top-priority target that ISR drones find easily and small or larger strike drones destroy fast. \"I think the era of tanks is probably over.\"\n- Even artillery hiding 40–50 km behind the front gets found and struck by drones. The taxonomy, prompted by Senra: FPVs are small, mostly manually flown quadcopters; strike drones like Helsing's HX-2 travel long distances and are \"quite often AI controlled.\"\n\n### 2. 100,000 strike drones would deter a land war in Europe\n- The tweet that \"got me into a bit of trouble\" — Reil stands by it: 3,000 tanks arriving against 100,000 strike drones that navigate without GPS and hunt tanks specifically get destroyed \"in a matter of hours.\" Senra flips it — how would you invade such a country? \"Very difficult\": you'd probably avoid those kinds of crewed vehicles and use smaller autonomous systems instead — \"this is basically where everything is going\" — drones, autonomous fighter jets, land robots.\n- Why autonomy is non-negotiable: Ukraine's thousands of skilled operators \"won't scale to the West and won't scale to NATO.\" Humans stay on or in the loop; AI flies. Status: \"reasonably solved\" — terminal guidance, the last roughly 2 km, and coordination software are the immediate capabilities being established, with the field now moving toward almost full autonomy.\n\n### 3. Jamming is increasingly circumvented, so the battlefield becomes machines versus machines\n- Two jamming problems: video links back to operators are jammed, so a drone must fly autonomously or use a cable — the latter is what is happening with some FPVs — while larger drones can fly themselves; GPS is jammed along \"pretty much the whole front line,\" so AI can navigate by recognizing its surroundings, like a pilot. The algorithms are now \"very robust... in daylight, at night, in the snow,\" even with terrain deformed by shell impacts.\n- This resolves the dinner argument with Senra from years ago (the Freedom's Forge thesis of out-producing the enemy): \"It's going to be a battle of material. It's machines versus machines.\" The extrapolation: \"it's going to be unethical to send humans on a battlefield where you know they're not going to survive\" — yet territory still must be retaken and defended, \"so you need robots on the ground.\"\n\n### 4. Founding Helsing in 2021, when defense was bucketed with pornography\n- Reil wasn't planning a second company post-Natural Motion — he was \"pretty low on fuel\" — and wouldn't have done it solo: defense founders juggle technology, politics, export law, classified information, and other complexities at once. He met Gundbert, then responsible for aerospace and defense at McKinsey, in a Berlin café, decided within 30 minutes, looped in Niklas, whose startup solved AI problems for legacy primes, and the three decided within days to start Helsing. It was founded in March 2021.\n- The climate: \"it was very unpopular.\" Some people close to the company had been debanked around then; Reil himself was not, while another person close to the company had difficulty getting credit because of its defense involvement. Banks grouped the defense of a country or democratic society \"in the same bucket as pornography and gambling and smoking.\" No VC would talk — \"we didn't even try in the end\" — so the €8M seed came from individuals and family offices.\n- The trigger was Crimea in 2014 — \"we all didn't want it to be true\" — plus the asymmetry that the US had Anduril and Palantir while \"Europe had nothing.\" \"It felt like if we don't start the company then no one else will.\"\n\n### 5. Daniel Ek's €100M and the decision never to sell\n- Three months after seed, Ek asked how much they needed. \"Ideally 100 million euros\" — no budget behind it, \"it felt about right... that's true for almost all funding rounds.\" Ek put in his own money via Prima Materia, before the invasion; Reil calls him \"closer to a co-founder than an investor.\"\n- On exits, categorical: \"we will never sell the company.\" The decision came months in: \"go full throttle and we either completely crash the thing or be super successful\" — nothing in between. Senra's supporting anecdote from AppLovin's Adam: stock down from roughly $28B to roughly $3B, borrowed roughly $6B for buybacks, made roughly $60B — \"I'm not running a $3 billion company.\" Reil: \"I think that's the only way to do it.\"\n- Senra's crusade against selling (\"Got acquired. Sorry to hear that.\") meets Reil's Tuscany theory: Europe is too pleasant to retire in, and once a few friends do it, it becomes socially accepted — \"it shouldn't be socially acceptable.\"\n\n### 6. Europe's missing ingredient is aggression — everything else is excuses\n- Mission before conviction: \"protect our democracies\" needs no offsite retreat to draft, and Spotify was the beacon showing that a globally consequential European company is possible. To regulation-blamers: yes, a German GmbH needs a notary — \"who cares?\" Defense compliance is \"like 100 times more than what a startup has to deal with... It still doesn't slow us down. You just have to set the company up properly.\"\n- The claim Senra flags as peak disagreeableness: ambition and aggression are \"95% of a company's success.\" The fix is role models: \"you just have to have a few very aggressive companies that point out that everyone else is basically mailing it in. And I think a lot of people are.\"\n\n### 7. The program-of-record blueprint: dark horse to prime in 28 months\n- An insight Reil thinks came from Gundbert, unlocked by the €100M: go straight for programs of record — large, budgeted government programs where one ideally has a direct prime relationship — which startups avoid because qualifying takes far too long for their proof-point cadence. With €100M, \"if the first two or three don't work... we still have enough money left.\"\n- The win: Germany's Eurofighter electronic-warfare upgrade. Many anti-aircraft radars had become software-defined — \"basically shapeshifting\" — so old detection approaches no longer worked; Helsing's deep-learning approach recovered the ability to interpret them, and with partner Saab it beat larger consortia as \"the dark horse,\" 2 years 4 months after founding. Reil thinks Helsing became the fastest company to go from founding to a program of record.\n- It became the template for everything since: \"we try to find programs of record that are large where we have a feeling that we can contribute something to it that's differentiated. That's basically our lens.\"\n\n### 8. A superhuman fighter pilot the pilots said couldn't exist\n- Two years ago Helsing set itself the task of creating a superhuman AI pilot for beyond-visual-range combat. Chief scientist Antoine Bordes joined from Meta, where he ran or co-headed Facebook AI Research. The combat is \"essentially chess in the sky,\" usually two-versus-two, involving radar, missiles, energy management, and re-engagement; it \"almost looks like ballet.\" Active pilots told Helsing this was the last bastion of air combat and that human intuition could never be replaced. \"That turns out to be [bull]... the AI is much, much better.\"\n- Why: reinforcement learning in the AlphaGo/AlphaStar mold, run in a simulator at 20–25,000x realtime, explores the search space in ways a flight-school-trained human cannot. Helsing regularly flies it against active Eurofighter, Rafale, and Gripen pilots.\n- The proof point: after winning in its own simulation and then being tested in an official simulator from a large prime, last year the AI \"took full control\" of a Saab Gripen over the Baltic Sea in three air-combat scenarios against a human-controlled Gripen. That was when Helsing realized \"there's something bigger happening.\" The system, called Centaur, now also works in swarms of fighter jets and drones, and \"it scales really well.\"\n- The meta-lesson on timelines: ten years ago everything was \"five years away\" and took 7–10; now \"when you think something is 5 years away, it's actually usually 2 years away.\"\n\n### 9. CA-1 Europa: buy the aircraft factory, take off from the autobahn\n- Partnering with a prime on an autonomous jet \"would have taken too long,\" so Helsing acquired an aircraft manufacturer outside Munich — runway, hangars, full manufacturing — to build the CA-1 Europa around Centaur. First flight next year, ambition \"many hundreds\" of jets, intent to \"leapfrog everyone in this category.\"\n- On the regulation Senra ribbed him about at Michael Dell's house: prototypes carry a safety pilot in the front because a crash lands on \"built-up Bavaria,\" not Mojave sand — accommodated without changing the design, like Waymo's early safety drivers.\n- The underrated vulnerability: exquisite jets like the Eurofighter and Rafale need airbases (the Gripen excepted) — obvious attack targets — while the CA-1 launches from any makeshift strip, motorway or highway: \"you can't bomb all of it.\" Senra offers the WWII Krupp parallel of centralized factories being found and bombed; Reil's verdict: \"certain governments and militaries are still in denial\" — though once demonstrations show the capabilities, procurement can move \"in record time.\"\n\n### 10. Good-enough hardware, smart software, factories designed not to be bombed\n- The unified-system thesis: design hardware to mass-produce and push complexity into software — simpler sensors whose accuracy AI recovers, looser manufacturing tolerances recalibrated in software — then win through \"pure mass\" and saturation attacks.\n- HX-2 numbers: last July–August production went from dozens per month to a 1,000 baseline and can now scale to 3,000, in one small factory — \"a completely different kettle of fish.\" Senra asks whether China is building roughly 100 million drones a year; Reil says \"not really\" and instead cites Ukraine building \"2 million or so\" of a different type: shorter-range FPVs. A strike drone's payload is 2–4 kg of explosive, or larger, with a range around 100 km.\n- The bottleneck isn't manufacturing speed but programs: a €1 billion German framework contract just closed, and \"they now need to come everywhere\" — Western countries plus some Asia. Reil said Germany's defense budget was €52B last year, originally projected, he thought, at €153B in 2029 and perhaps now above €180B. Since there will most likely not be more soldiers, he expected most of the money to go to equipment.\n- Sovereignty via \"resilience factories\": Helsing hardware built on the customer's soil, supply chains localized to each country's sensitivity (China excluded, allies welcomed), plants distributed across a continent or underground \"so that they are harder to hit.\"\n\n### 11. Deterrence ethics — and the killer-robot questions nobody asks anymore\n- The moral architecture: minimize lives lost, ideally to zero, which means no conflict at all — hence deterrence, \"a credible posture that makes the cost calculus on the other side such that they don't want to attack because it'll be too devastating for them.\" The best outcome for Helsing is literally that nothing happens.\n- Whether a machine should decide over life and death \"if it is more accurate than a human\" is, per Reil, a legitimate question with \"no easy answer.\" His worry is who stopped asking: before the Ukrainian war, journalists constantly probed killer robots, red lines, and humans in the loop — \"these weren't annoying questions\" — and now \"no journalist ever asks those questions.\" Helsing runs ethics workshops and deployment guides \"out of an inner discipline.\"\n- On pacifist pushback — including in Germany, whose \"entire postwar identity was never again\" (Senra channeling Ek): \"we want to have peace too. We just think you have to achieve it by not letting a war happen.\" On human nature: the appetite for conflict is partly cultural, partly \"hardwired,\" probably genetic in part of the population — you likely can't eliminate the outliers, so contain them and \"look very prickly.\"\n- Learning Helsing drones were deployed in Ukraine: \"I personally found it very difficult... it became a more serious company,\" the light-heartedness of games and SaaS gone. \"We don't take it lightly.\"\n\n### 12. Talent density above everything: the hiring machine and the firing math\n- Reil's confessed NaturalMotion mistake — not taking hiring and performance management seriously enough — became Helsing's founding design: the first contract, signed even before incorporation, was with a recruitment agency, and the three founders reviewed recruiting KPIs weekly for the first year.\n- The load-bearing insight: hiring and firing are linked. Managers keep bad fits because refilling takes nine months — \"at least eventually they'll get the job done... and that's not good enough.\" Pipeline pressure makes moving people out \"a low opportunity cost thing,\" and quarterly calibration reviews hold managers accountable for their ratings — \"that doesn't sound like a three, that sounds like a two\" — exposing who protects underperforming teams.\n- His interview method: no standard questions, roughly 10 minutes reading for \"effortlessness\" — not trying to impress, deep knowledge, \"still curious and have high agency.\" He recalls no such miss at Helsing, but three at NaturalMotion (\"I still know the names\") by people faking it with fake facts. Senra's counterpunch on delegation — Elon personally approved SpaceX's first 3,000 hires; Bending Spoons runs 50 HR engineers correlating CV signals with multi-year performance — draws agreement on the principle: \"it compounds over time.\"\n- Culture is downstream of density: at a company get-together, the team spontaneously rewrote Helsing's values — \"whatever it takes\" (people back from the Ukrainian front line, 48 hours without sleep) and \"the truth is in the field,\" because systems must work in combat, not government tests. And \"the power law in talent is extreme\": Helsing makes exceptions for its 10-out-of-10s, owed \"gratefulness\" for turning projects \"from red to green.\"\n\n### 13. From walking neural nets in 1999 to CSR Racing — the outsider's throughline\n- The origin Eureka: a 1999 complex-systems PhD working with fully recurrent neural networks and genetic algorithms modeled on biological central pattern generators, hooked to early physics engines such as MathEngine and Havok — overnight, \"I literally had this two-legged creature walking over the screen, fully physically simulated.\" That became NaturalMotion: Endorphin at $50k a license in Lord of the Rings, Troy, Spider-Man, and Harry Potter; real-time technology in Rockstar's GTA IV, GTA V, Red Dead Redemption, and others; then making games itself — CSR Racing, \"the biggest game of all time on the iPhone at the time,\" delivered the hockey stick.\n- Senra fishes for the connective theme between his two companies; Reil's answer is outsiderness: never feeling part of the biology or gaming peer group made contrarianism easier — pricing games at 99 cents to maximize revenue when the industry insisted on $5, going first-big in free-to-play 3D. \"It would be much harder for us to build Helsing if the backgrounds had already been in defense companies. It's easier to be an outsider.\"\n- The closing salvo on LinkedIn/X culture: \"commentary is entirely worthless... it has a net negative benefit,\" since platforms reward negative commentators over creators. Prescription: \"You either do something about it or you don't talk about it\" — and founders \"have the biggest lever\" to change society, which is why so few of them are posting.\n","transcript_md":"David Senra\n\nLet’s start with what we were just talking about before we started recording. What’s your opinion of the future of tanks and battleships?\n\nTorsten Reil\n\nI think there’s a big change happening, and you can already see it in Ukraine. Tanks are barely being used anymore. The same is true for artillery systems. If you get yourself into a tank, your survival rate is very short.\n\nApparently, the time that a Russian soldier survives at the front now is about 20 to 30 minutes. You can argue about whether that’s correct, but it’s probably roughly right in terms of ballpark. It probably gets even worse if you’re in a tank. If you put yourself into a tank, you are one of the high-priority targets for anyone on the other side.\n\nThose targets can be found pretty easily now with drones. They can be destroyed very quickly with small drones, or with large drones if they’re a little bit further away. So, I think the era of tanks is probably over. For artillery systems as well, if they try to hide 40 to 50 kilometers away from the front to shoot, you can still find them with drones and attack them with drones.\n\nAgain, they’re high-priority targets. So, I think we’re seeing a big change right now in warfare. Those lessons have been pretty apparent, I think, in Ukraine over the past 18 months or so. People are starting to understand them now.\n\nDavid Senra\n\nWhat’s killing the Russian soldiers so quickly? Is it all drones?\n\nTorsten Reil\n\nIt’s drones. It’s the ability, first of all, to see them with ISR drones and, secondly, to strike them with either FPV drones or larger strike drones.\n\nDavid Senra\n\nWait, what’s the difference between those 2 drones?\n\nTorsten Reil\n\nFPVs are basically smaller drones—quadcopters that, most of the time, are controlled manually, like a remote-controlled plane. Strike drones often travel over longer distances and, like the HX-2, which is our drone, are quite often AI-controlled.\n\nDavid Senra\n\nYou had a tweet that was interesting, and maybe this can tie into the dinner that we had a few years ago, where we had this fierce argument. I left thinking that we hated each other, and then we became friends. You said something like, “100,000 of your drones could prevent a land war in Europe.” What does that mean, and why do you believe that?\n\nTorsten Reil\n\nI said this maybe a year ago, and I got into a bit of trouble for it because apparently it was deemed to be provocative. I still think that’s the case. The deterrence that you get out of having a very large number of strike drones at your border is pretty strong.\n\nThe reason is that if you turn up with tanks on the other side—let’s say you turn up with 3,000 tanks—100,000 strike drones that are able to navigate without GPS and find their targets, specifically focusing on tanks, will destroy those tanks, I think, in a matter of hours. It’s very hard to defend against. If you were on the other side and wanted to invade a country that had 100,000 of your drones, how do you do that?\n\nDavid Senra\n\nI think it’s very difficult. Using tanks would be extremely difficult because, again, they’d be targets immediately. You’d probably want to not use those kinds of vehicles. You’d probably want to use autonomous systems—smaller autonomous systems.\n\nThis is basically where everything is going. You have autonomy now with drones. You have autonomy coming up with autonomous fighter jets, and you also have it coming up with land systems, in particular robots.\n\nWhat do you mean by autonomous drones, though? Are they all run by AI?\n\nTorsten Reil\n\nThey will be. At the moment, especially in Ukraine, a lot of drones are still remote-controlled. You have thousands of very skilled drone operators. That won’t scale, though, particularly in the West, and it won’t scale to NATO.\n\nIf you really want scaled drone usage, they have to fly autonomously. You want to have a human surveying them, looking at what they’re doing, and being on or in the loop, but the drones themselves need to be AI-controlled and autonomous. It’s the only way to achieve true mass, and it’s also the only way to make it work pragmatically, because we won’t be able to train thousands and thousands of drone operators. We need AI to control them.\n\nDavid Senra\n\nIs that a solved problem yet, or have you still not figured it out?\n\nTorsten Reil\n\nIt’s reasonably solved. I think the need so far has been, first of all, to make the drones work for terminal guidance, which is basically the last 2 kilometers or so, and then to create the software to coordinate them all. Now it’s going to move toward almost full autonomy, again always with a human in the loop.\n\nDavid Senra\n\nI want to go back to the conversation we had a few years ago, when we were having dinner at Daniel Ek’s house. Your proposition to me was essentially that you’re just not going to be able to have humans on the battlefield anymore, because it’s going to be about who can produce more material.\n\nI think we discussed this book called Freedom’s Forge, which I’ve read twice. Essentially, it’s about the United States converting its industrial and commercial power from making products that people would buy to making war materials, and mass-producing more than the opposition had. That helped tilt the balance of the war in its favor.\n\nI thought you had a very similar perspective: that’s what we would have to do. The winners of future wars are just going to have to mass-produce more drones, more weapons—more of everything.\n\nTorsten Reil\n\nI think it’s going to be a battle of material. It’s machines versus machines. The reason for this is that autonomous mass, as we call it—the number of drones that can fly in a coordinated way—is getting better and better, and it’s going to be very hard for humans to survive on that battlefield.\n\nAgain, that’s already the case in Ukraine. You can’t really enter the battlefield as a human and survive for very long. Whether you’re in a vehicle or on your own, it’s very hard to survive. That’s going to get worse and worse because, at the moment, the drones attacking you are usually remote-controlled. Once they’re autonomous, it’s going to be even more dangerous.\n\nThose drones usually don’t care about jamming anymore. For FPVs, that’s because they’re connected through fiber-optic cables. For larger drones, it’s because they’re already autonomous and don’t have to worry about GPS, because they have GPS-free navigation. They can detect targets and engage once you’ve allowed them to do so autonomously.\n\nDavid Senra\n\nWait, explain how you get around the jamming. I don’t understand that.\n\nTorsten Reil\n\nThe problem with jamming is twofold. First of all, if you want to control your drone from afar, you need a video signal back, and usually the battlefield is jammed. So, you don’t get the video signal back. That means the drone needs to fly autonomously, or it needs to be connected with a cable. That’s what’s happening right now.\n\nThe second problem is GPS jamming. Pretty much the whole front line is GPS-jammed, so you can’t use your GPS receiver. You need to find a different way to navigate and know where you are. You either do that with a human trying to recognize where they are, almost like a pilot of a plane does, or you do it with AI that tries to do the same thing.\n\nThose algorithms are now very robust. They work in daylight, at night, in the snow, and even with deformed terrain, which obviously happens quite a lot with shell impacts. That technology already exists.\n\nThese drones are getting better and better, and at the same time, if you want to have humans on the battlefield, I think it’s going to be very difficult. I think the trend is going to be—and people are realizing this—that it’s going to be unethical to send humans onto a battlefield where you know they’re not going to survive. At the same time, you still have to retake territory and protect and defend territory. So, you need robots on the ground. If you extrapolate from where we are right now, I think that’s where we’re going to go.\n\n### Why Torsten Founded Helsing\n\nDavid Senra\n\nI want to go into the way you’re running Helsing right now, but I think the place to start is why you decided to found the company to begin with. We can give a quick overview. You came from gaming, right? You sold a company for half a billion dollars, and you were angel investing, looking for your next thing. Did you know for a fact that you were going to found another company after your first sale?\n\nTorsten Reil\n\nNo. After I left NaturalMotion and Zynga, I didn’t think I was going to start another company. I was pretty exhausted, to be honest. I’d been doing it for 15 years, and I was pretty low on fuel in my tank.\n\nIt didn’t take very long to recover—about 3 months, maybe—and I felt the energy was back. But then it took another 2 or 3 years before I thought, “Actually, I want to start something again.” Honestly, I wouldn’t have done it had I not met my 2 co-founders, Gundbert and Niklas.\n\nEspecially with what we’re doing in defense, it’s so complex. Being a founder is already lonely as a single founder, but in defense, the companies are so complex. You have to think about so many things at the same time. You obviously have to deal with technology, but you also have to deal with politics, export law, and a whole bunch of other things. It’s very difficult to do it as a single founder.\n\nSo when I met my two co-founders, we had all independently thought that there needed to be a new defense company in Europe.\n\nDavid Senra\n\nHow did that conversation even start?\n\nTorsten Reil\n\nBack then, I invested in Niklas’s company in Munich, which was working on solving AI problems for some of the legacy primes in Europe. I wanted to find out more about defense because I’d invested, but I didn’t know that much about it. Someone recommended I meet up with Gundbert, who at the time was at McKinsey, responsible for aerospace and defense.\n\nWe met up in a café in Berlin and hit it off. Within 30 minutes, we thought we should start this company together. We went to see Niklas, who had the same feeling, and within a few days or so, we decided to start Helsing. This was a very high-conviction decision, and there was never a doubt. That’s still the feeling of the company now: the company moves very fast, and a lot of it is based on quickly settling on founder conviction in terms of what we do—that we need to go into certain areas, certain territories, et cetera.\n\n### Founder Conviction & Betting on European Defense\n\nDavid Senra\n\nHold on. I don’t want to move on from the conviction yet, because the year you’re having these conversations—what year is this?\n\nTorsten Reil\n\nThat was the end of 2020, beginning of 2021.\n\nDavid Senra\n\nOkay. I can’t think of a more unpopular company to start at that time than a European defense tech company.\n\nTorsten Reil\n\nYeah, it was very unpopular. We founded the company in March 2021, and it wasn’t popular in our social circles. It caused all kinds of problems with banks. Although this has gotten better now, obviously, back then it was quite unpopular, and it was hard to raise money.\n\nDavid Senra\n\nWait, it caused problems with banks?\n\nTorsten Reil\n\nThere was a risk, and some people who had been close to our company had been debanked around that time. I wasn’t, but I know another person close to the company had difficulty getting credit because we were involved in defense, and this is still the case now. It hasn’t completely disappeared. I think ESG requirements are still there.\n\nBack then, the defense of a country, of a democratic society, was grouped by banks in the same bucket as pornography, gambling, and smoking because it was seen as weapons, and weapons are always bad. I empathize with the gut feeling on this; it’s just that when you think it through, obviously it doesn’t make sense.\n\nThis is also why we started the company. It felt like if we didn’t start the company, no one else would. We were specifically worried about Europe’s reaction to Crimea in 2014—the invasion and the taking of Crimea. I think all of us didn’t really want to think that this was the beginning of something bigger. We all didn’t want it to be true. But if you looked at what the ambitions were, it was pretty clear that something else was going to happen.\n\nWe were concerned that Europe was going to be very dependent on other territories for its defense, and on other countries. The U.S. had companies like Anduril and Palantir that were very forward-looking defense companies, with a big focus on AI, cheaper hardware, abundant hardware, and the idea of mass manufacturing. But Europe had nothing.\n\nWe felt we had to do this, and we had to prove that we could build something like this in Europe, although with a slightly different flavor. That’s why we started it. It was very difficult to raise money in 2021 for defense. No VC wanted to talk to us, and we didn’t even try in the end. We just raised money from individuals and family offices, and then we were lucky that Daniel put in €100 million of his own money very early on. That allowed us to scale the company faster than we otherwise would have been able to.\n\nDavid Senra\n\nThis is what I always read about you guys: “Yeah, and they got a $100 million investment from Prima Materia.” That’s his own money.\n\nTorsten Reil\n\nYeah.\n\nDavid Senra\n\nThat’s a high-conviction bet really early. This is before the invasion of Ukraine.\n\nTorsten Reil\n\nYeah.\n\nDavid Senra\n\nThere was a seed round first, right? Didn’t you guys raise, like, €8 million or €10 million or something like that?\n\nTorsten Reil\n\nYeah, I think it was €8 million in March 2021.\n\nDavid Senra\n\nAnd he was in that seed round?\n\nTorsten Reil\n\nYeah, he was in that seed round, too.\n\nDavid Senra\n\nAnd then you guys had this conversation where he asked, “Well, how much do you need?”\n\nTorsten Reil\n\nYeah, about 3 months later.\n\nDavid Senra\n\nAnd your response was—\n\nTorsten Reil\n\nIdeally, €100 million.\n\nDavid Senra\n\nHow’d you come up with that number?\n\nTorsten Reil\n\nI didn’t really do a specific budget for it, to be honest.\n\nDavid Senra\n\nBut it felt about right.\n\nTorsten Reil\n\nIt felt about right. That’s true for almost all funding rounds, isn’t it? You kind of eyeball it because so many things change so quickly in a company. It’s very hard to find an exact number.\n\nBut Daniel had the same level of conviction that Gundbert, Niklas, and I had, and I think he’s a similar-conviction person. It doesn’t take very long, and then the conviction is very deep, and it makes everything easier. It made all the conversations easier. We didn’t really have to struggle to come to a deal; it just happened really quickly.\n\nDavid Senra\n\nYou consider him more like another co-founder than an investor?\n\nTorsten Reil\n\nYeah, I think he’s closer to a co-founder than an investor because he’s been with the company from the beginning. I knew Daniel before. He knew about the idea from the beginning. It was a big commitment for him to also be publicly seen investing in this kind of company.\n\nObviously, what happened was that only 11 months after we started the company, the war in Ukraine happened and public sentiment shifted. It’s now easier, and a lot of people are investing in defense, both in the U.S. and in Europe. But at the time, it was extremely unpopular.\n\nDavid Senra\n\nI think this is one of the things I most admire about you: this idea of founder conviction. I love that you’re—I’ve heard you say this before—it’s like the limiting factor is not capital. It’s actually ambition.\n\nTorsten Reil\n\nYeah.\n\nDavid Senra\n\nOne thing that I’ve talked to a bunch of people who have been on this podcast about is that I try to encourage them: stop selling your goddamn companies.\n\nTorsten Reil\n\nYeah. Your whole thing is, like, stop selling your company for $100 million and then buying a house in Italy and becoming an investor. We need more people. We have enough people in [__] finance, dude. You need more people doing [__].\n\nI completely agree. I have this theory that the reason this happens so often in Europe is partly because Europe is such a nice continent to retire in. You can buy a house in Tuscany, and you can fly there from Scandinavia. You’re there in about 2 hours, and you can live an amazing life. Once a few of your friends start doing that, then it’s kind of socially accepted. But it shouldn’t be socially acceptable that you retire as a founder who’s been successful and can start another company.\n\nDavid Senra\n\nMy friends who sell their companies—I say this jokingly, but there’s a lot of truth in it—they’re like, “Got acquired.” And I’m like, “Sorry to hear that.”\n\nTorsten Reil\n\nYeah, exactly.\n\nDavid Senra\n\nEntrepreneurship is the best game in the world. You took yourself out of the game and put yourself on the [__] sidelines to do what? Write checks.\n\nTorsten Reil\n\nNo, I agree. It’s easy for me to say because I sold my previous company as well. It may be too early, although I’ve been doing it for a while. As I said, for a while I thought I wasn’t going to start another company because building companies is exhausting, and sometimes you need to recharge your batteries.\n\nBut once you’ve done it, I think people are also afraid that they might not find the right mission, or that they might find that the second company isn’t as successful as the first one. That is a risk. I thought about that, too: What if the whole thing just doesn’t work at all? But I think that’s a risk you have to take.\n\nDavid Senra\n\nI just talked to Scott Wu from Cognition, and they’re throwing all kinds of crazy money at him. This is huge—a million people are trying to acquire him as a person or the entire company. His whole thing is, “Yeah, but this is the biggest opportunity of my life. I couldn’t live with myself if I didn’t at least try. I’m fine failing.”\n\nAnd then he said, “I’d be salty as hell,” meaning he’s super competitive. To be removed from the game completely, as some kind of piece on somebody else’s chessboard, is just intolerable to somebody like that.\n\nTorsten Reil\n\nYeah, no, totally. For us with Helsing, it’s the same thing. We made a decision probably 4 and a half years ago, just a few months into the company, that we were going to go full throttle. We either completely crash the thing or be super successful, but we didn’t want to do anything in between.\n\nWhat we’re doing right now is trying to keep our foot on the gas really aggressively because we want to build a company that really matters, and we will never sell the company. I have a funny story about this, and then we’re going to get back to Helsing. Adam, the founder of AppLovin, was sitting in the chair that you’re in now, and he tells this hilarious story.\n\nHe takes the company public—it’s valued at $28 billion or something like that. Every day for 200 days, the stock just goes down and down and down, and it gets down to about $3 billion.\n\nDavid Senra\n\nHe's like, “This is either a multiple-hundred-billion-dollar company or it's nothing.” So then he winds up borrowing—he goes into debt by like $6 billion—and does stock buybacks, then winds up making like $60 billion on that trade. He's like, “I'm not running a fucking $3 billion company. This is either a big company or nothing, and I'm not playing this little game.”\n\nTorsten Reil\n\nI think that's the only way to do it.\n\n### Helsing's Mission: Protecting Democracies\n\nDavid Senra\n\nSo, the identification of what was missing in Europe, which leads to this mission that we need a European defense tech company: Does the mission come before the conviction, or does the conviction come before the mission?\n\nTorsten Reil\n\nThe mission comes first, and the mission is to protect our democracies. I always say it's a real privilege. We talk to our new starters every month, and we always say to them, “It is a real privilege to work in a company where the mission is so easy: We protect our democracies.” That's what we're here for.\n\nLots of other companies have to spend time and go on an offsite retreat to come up with some kind of mission paragraph that everyone eventually agrees on. We don't have that. It is to protect our democracies. We can point to where we're doing this right now with what we're doing with our deployed drones, for example, in Ukraine, and it makes everything easier.\n\nWe obviously never expected the war in Ukraine to happen that quickly, but now that we are deployed, it's very concrete. This mission is a real thing. Then the conviction comes that these kinds of companies need to exist and that they need to be built in Europe.\n\nAnother reason why we are building Helsing is that we wanted to build a big company in Europe. We honestly looked at Spotify as a beacon, as an example of a large company you can build that has global ambitions, but also has global impact and is up there with other big companies from the US and other countries. We wanted to build something like that, too. We wanted to show that you can do this, and you can do it in what is a heavily regulated area like defense.\n\nA lot of people find excuses for why they can't build a big company. It's very often regulation, bureaucracy, or not enough funding. None of that is true. They're all excuses.\n\nDavid Senra\n\nYou have a great line about this. I love what you said: The excuse is that people are way too concerned about the regulatory environment and the tax environment. What's missing is that you need aggressive founders with large ambitions. It's just aggression that's missing.\n\nTorsten Reil\n\nExactly. It's strategic ambition. You have to think about what you want to be, and it needs to be big. Then you need to be truly aggressive. A lot of founders, I think, in Europe have lost that aggression, but it's easy to find it again.\n\nYou just need role models, like we use Spotify as a role model. Once you have them, that's what good looks like, and everything below that is failure. It's underestimated how much of a difference aggression and ambition make.\n\nDavid Senra\n\nSay more about that.\n\nTorsten Reil\n\nI think it's 95% of a company's success; everything else, like I said, is excuses. Regulation doesn't matter. Yes, sure, it takes longer to start a GmbH in Germany. Who cares? Yes, you have to go to a notary. Who cares? We spend a couple more days starting the company. It's not a problem.\n\nThe amount of regulation that we are subject to—and that other defense companies are as well—with regard to the certification and qualification of our systems, export law, and a whole bunch of other things, including confidentiality and classified information, is like 100 times more than what a startup has to deal with in terms of bureaucracy. You can still deal with it. It still doesn't slow us down. You just have to set the company up properly.\n\nDavid Senra\n\nYou're probably one of the most disagreeable people that I know. Have you always been that way?\n\nTorsten Reil\n\nYeah, you said that to me before, and I've been trying to reflect on it. I don't think I am.\n\nDavid Senra\n\nYou just demonstrated it here. We have more regulation. Who gives a fuck? Stop making excuses. It's a superpower for founders. Obviously, I know it's not a pejorative when I say that.\n\nTorsten Reil\n\nI just don't like it when people make excuses. These things happen very quickly. You just need to have a bubble of people coming together and all reinforcing each other with these kinds of excuses, and then they have a content life.\n\nSadly, we have a continent where we don't have breakout companies because people just lack the aggression. But again, I am very optimistic. There are quick fixes. You just have to have a few very aggressive companies that point out that everyone else is basically mailing it in. And I think a lot of people are.\n\nDavid Senra\n\nDaniel and I talked about this idea of how many prime movers there are in the world—the people who are inventing the technology that we're all using and creating the companies. Power laws rule everything around us, and it's just not that many. If you just have a handful, you can literally change the trajectory of an entire country, an entire world, whatever the case may be.\n\nWhen you raised the $100 million, what did you see as your entry point into doing this? Was the idea, “Hey, we're going to do what you just said: cheap, simple hardware, and then all the innovation comes in software”? Is that the way you thought about it?\n\n### Programs of Record & Winning the Eurofighter Contract\n\nTorsten Reil\n\nActually, the biggest insight came—and I think it came from Gundbert—when we said, “We're just going to go for programs of record,” which, back then, as a small company, was almost impossible and extremely risky.\n\nDavid Senra\n\nExplain what a program of record is.\n\nTorsten Reil\n\nA program of record is an official government program that's in the budget, usually quite large, where you ideally want to be the prime. So you have the direct relationship with the government rather than working as a subcontractor through another company, and that's extremely hard to break into.\n\nThe reason most companies didn't do it back then was that you got funding and needed to generate some revenue and proof points reasonably quickly. That's almost impossible to do with a program of record because it takes so long to get one and to qualify for one.\n\nBut because we had the $100 million, we thought, “Okay, now we can take those risks. If the first 2 or 3 we go for don't work, we still have enough money left.” We felt it was the only way to build a defense company reasonably quickly.\n\nSo we applied for a big program of record for the Eurofighter electronic warfare upgrade.\n\nDavid Senra\n\nWhat is that?\n\nTorsten Reil\n\nIt's in Germany. Electronic warfare capabilities are essentially about detecting radars—the types of radars, in this case antiaircraft radars, that try to take you down—detecting their intent, and so on.\n\nThe situation had become more difficult for Europe because a lot of those radars had become software-defined, which means that they basically are shapeshifting, and the old approaches didn't really work anymore. We came in with an AI-based, deep-learning-based approach that recovered the ability to interpret those radars.\n\nIt was a big technology jump, and we managed to win that program of record together with our partner Saab against a lot of other consortia. We were the dark horse, but we ended up winning it.\n\nHow big was the probability? I don't know, but there was a high probability that we'd fail. We went for it because we thought we could afford to, because we had enough money in the bank. This happened 2 years and 4 months after founding the company. I think we became the fastest company to go from founding to a program of record.\n\nDavid Senra\n\nHow much of your previous work at Oxford and the stuff you were doing at NaturalMotion did you bring into Helsing?\n\nTorsten Reil\n\nA lot of it was, I think, team building, fundraising, prioritization, and pattern recognition that I had from NaturalMotion. When I sold the company and left it, I was reasonably pleased with myself. I thought, you know, this all went quite well in the end. But it was only in the 2 or 3 years afterward, when I started to reflect on some of the things I'd done and how I'd built the company, especially when I started to angel invest, that I realized I had actually screwed up a lot of things at NaturalMotion that I should have done better.\n\nI didn't deal with performance management the way I should have. I definitely didn't take hiring seriously enough—the general hiring of great engineers. I also didn't think enough about how to think about your leadership team, how you grow your leadership team, and how you bring the right executives in at the right time. It kind of worked at NaturalMotion, but we probably could have done a lot of things better.\n\nA lot of that pattern recognition we were able to use at Helsing, particularly in the first few years. But it's also true now that Helsing is at a point that is uncharted territory for me as well. The company is a lot bigger. I think we're about 1,300 people, much bigger than NaturalMotion ever was, so we now have to think through things either from first principles or by working with people who have already built big companies, like Daniel.\n\nThe other part is that we were lucky that the 3 of us as founders had 3 very different skill sets. I had already started and sold a company, so I think I knew how to fundraise. Gundbert was very close to the various forces and the military, and also understood the political processes really well. Niklas's background was on the AI and engineering side, which is what we really needed to first of all win this program of record.\n\nDavid Senra\n\nAnd this is what I was getting at. That was the reason I asked that question: how the hell do you go from founding to a program of record faster than anybody else? How did you even know you could solve this problem?\n\nTorsten Reil\n\nIt was just a feeling, I think. In particular, Niklas and Gundbert pushed that. Gundbert had a strong conviction that this was a really worthwhile program of record for us to have, and Niklas had an intuition as to how we could do it differently from anyone else.\n\nDavid Senra\n\nWhere did Niklas get that intuition?\n\nTorsten Reil\n\nI don't know. I think it's just that he'd already been in this space with his previous company, so he already had a feeling for this kind of thing.\n\nDavid Senra\n\nWhat was his previous company doing?\n\nTorsten Reil\n\nIt was solving AI problems for legacy primes. So he already had a feeling for this kind of thing, and if you combine that with Gundbert's insight as to what programs of record worked, it honestly became the blueprint for how we approached pretty much everything.\n\n### Building a Superhuman AI Fighter Pilot\n\nNow we try to find programs of record that are large, where we have a feeling that we can contribute something differentiated. That's basically our lens.\n\nDavid Senra\n\nCan you give me an example of that?\n\nTorsten Reil\n\nAnother example is the autonomous fighter jet that we are building right now. It's called the CA-1 Europa. We went into it and invested a lot of money in it, knowing that Europe didn't have anything like it, that we felt this was going to be a big category, and that we could contribute something to it.\n\nThe reason for that was that we started almost like a science experiment. We have a great chief scientist, Antoine Bordes, who joined us from Meta, where he ran or co-headed Facebook AI Research. We set ourselves a task 2 years ago: to create a superhuman fighter pilot for beyond-visual-range combat.\n\nDavid Senra\n\nHold on. You've got to slow this down. What did you say? Superhuman? Say that thing again.\n\nTorsten Reil\n\nIt's a superhuman fighter pilot.\n\nDavid Senra\n\nSuperhuman fighter pilot, but there's no pilot in there?\n\nTorsten Reil\n\nNo, it's an AI. It's a neural network.\n\nDavid Senra\n\nOkay.\n\nTorsten Reil\n\nIt's for beyond-visual-range combat, the most advanced type of air combat. It's essentially chess in the sky. Usually it's 2 versus 2, quite far away from each other. These are fighter jets, and they use radar. They use different types of radar, and they switch the radar on and off. They use missiles. Obviously, you have to go offensive and defensive, try to bleed off the energy of missiles that are following you, and have a feeling for when you can reengage.\n\nIt's extremely complex. It almost looks like ballet. It's quite beautiful to watch, actually, when you look at it in the tactical views. We've been told by active pilots that this is the last bastion of air combat. Nothing will ever replace the intuition that a human has in the sky to read the situation, to have a feeling for how to engage, and what types of weapons to use.\n\nThat turns out to be bullshit. The AI is actually much, much better. The reason is that it's able to explore the search space much more efficiently than a human can. A human learns from a textbook and in flight school, but they can't explore the search space.\n\nWe use basically the same approaches as DeepMind used for AlphaGo and AlphaStar, using reinforcement learning. We have an extremely high-performance simulator that runs 20,000 to 25,000 times real-time speed, and that allows us to explore the search space with reinforcement learning much more efficiently.\n\nThat means we can train a neural network to be better than an active-duty fighter pilot. We regularly fly against Eurofighter pilots, for example, or Rafale pilots or Gripen pilots.\n\nDavid Senra\n\nSo wait, you were running tests with your autonomous fighter jet against a human-powered—\n\nTorsten Reil\n\nPilot. So we used to do it in simulation only. First it was only our simulation, and we won. Then it was an official simulator from a big prime. Last year, we flew a real fighter jet. We flew the Saab Gripen over the Baltic Sea.\n\nWhen I say “we,” our AI did that and took full control of the plane in air combat against another Gripen that was human-controlled. We flew 3 air-combat scenarios over the Baltic Sea and proved it.\n\nThis was also the point where we realized that something bigger was happening: this actually works, and it works in the real world. We developed the system and called it Centaur. It's now able to work in swarms as well, by the way—not just 2-versus-2 swarms of fighter jets, but now swarms of drones.\n\nDavid Senra\n\nSwarms of fighter jets?\n\nTorsten Reil\n\nAnd it scales really well. It's extremely powerful. That gave us the conviction that crewed, manned fighter jets were going to be very difficult to deploy in the future, because you'll end up facing these swarms of autonomous fighter jets with superhuman capabilities. It's not science fiction.\n\nOne observation of what's happening at the moment, not just in defense AI but in AI in general, is that things happen faster than you think. Back in the old days, 10 years ago or so, everything was always 5 years away. That was always the number people used, whether it was a particular medical breakthrough or a robotics breakthrough.\n\nThat's all shrunk now. The 5 years usually ended up being 7 to 10 years, by the way. The same was true of autonomous driving. It's now the exact opposite: when you think something is 5 years away, it's usually now 2 years away.\n\nSomething has changed, and we felt that with Centaur as well. We couldn't believe how well it worked. To this day, I always feel like it seems a bit like science fiction, but that gave us the conviction that we needed to be in this space.\n\nBecause it's usually so slow to work with primes, we could have partnered with a prime and asked, “Do you want to do an autonomous fighter jet? We provide the AI, you provide the jet.” But it would have taken too long. So we decided to acquire an aircraft manufacturer outside Munich, with its own runway, hangars, and full aircraft-manufacturing capabilities, and build our own fighter jet.\n\nThe aircraft is now called the CA-1. We put Centaur on it as the core technology component. It's going to fly for the first time next year, and we want to show that you can really leapfrog everyone in this category.\n\nDavid Senra\n\nWe were having this conversation—you, me, and Michael Dell—at Dell's house a few months ago. You told us about this acquisition, and then Michael said, “Can we come see it?” I said, “I'll come with you, but I don't want you to shoot down Michael's plane as we're coming in.”\n\nBut you told us about this funny regulation where, even though it's going to be an autonomous fighter jet, you still have to have a guy in it, don't you?\n\nTorsten Reil\n\nSo eventually, no.\n\nDavid Senra\n\nRight now, yes.\n\nTorsten Reil\n\nYeah. I mean, we don’t have the advantage of the Mojave Desert, for example, where you can fly those planes and be completely autonomous because, if it crashes, it doesn’t really matter. It crashes into sand. If you crash where we fly, you crash over built-up Bavaria, which is not ideal. So we have a human optionally in the plane for the first prototypes. Eventually, that’s going to change, obviously, and the pilot is a safety pilot, the same way you’d have safety drivers for the early rollouts of Waymo and other companies.\n\nDavid Senra\n\nThe fact that you have this regulation at the moment and need a person has to change the design of what it looks like, correct?\n\nTorsten Reil\n\nNo, we found a way to accommodate it without changing the design. It’s the same system. By the way, it’s another example of where you can beat on the EU—or Europe, or in this case Germany—for having regulations, but you just find a way to deal with it. This is the way to deal with it.\n\nDavid Senra\n\nYou just have a pilot sitting in there like this. Is he in the back? Where is he?\n\nTorsten Reil\n\nThey’re in the front.\n\nDavid Senra\n\nOkay. What you just described is very interesting. You said it’s like chess in the sky, and then you said it was ballet. So the way you see the short-term future is that it’s essentially going to be unmanned. I want to get into the morality—the way you look at the morality of what you’re building, and the way it would be immoral to send a human into the battlefield because you’re going to get swarmed by drones. It’s going to be immoral to send a pilot into the sky. Is that the same?\n\nTorsten Reil\n\nYeah, it’s the same thing. I mean, there are 2 reasons. First, it’s more powerful to use autonomous systems. They will just be superior because the AI is smarter than humans for this use case. Secondly, as a result, it will be immoral to send someone out because you have to assume that the other side will eventually have those systems as well.\n\n### Deterrence, Autonomous Weapons & the Morality of Defense\n\nDavid Senra\n\nCan you explain how you view the morality of what you’re building, and how you may view other people’s interpretation of essentially banning the creation of new weapons or new defenses? You’ve thought a lot about this. I want a complete overview of how you think about the morality of building the company you’re building.\n\n### How Torsten Hires Exceptional People\n\nTorsten Reil\n\nThe underlying goal has to be that you minimize the lives lost in general, and ideally you get it down to 0. The way to get lives lost down to 0 is to not have any kind of conflict at all. That’s why deterrence is so important. This is also where—I mean, this is a very old, obviously, argument, but it still comes up all the time—people don’t understand that you have to be able to deter in order to not be a target that looks valuable or achievable for the other side.\n\nWe set up the company to deter. We never set it up, honestly, to have the systems deployed at scale in Ukraine, which is what we have right now. We’ll do it, obviously, because we’ll have to do it, and it needs to make a contribution. But the idea in general is to avoid conflict altogether, and you do that by being able to deter—by having a credible posture that makes the cost calculus on the other side such that they don’t want to attack because it’ll be too devastating for them. That is the first thing.\n\nIf you have to go into a conflict, you want to do it in such a way that you minimize the lives lost on your side. Again, this is where, especially right now—and it’s going to get worse in the future—you need to use autonomy because you cannot send humans into tanks, fighter jets, or even as infantry onto the field. That’s the underlying theme.\n\nThen there’s another category where you need to think about the way you deploy your software and what, as a society, we’re happy with in terms of the deployment of your hardware and software in terms of autonomy. Are we okay for a machine to make a decision over life and death if it is more accurate than a human, or do we say, as a society, no, a machine should never make that decision? These are legitimate conversations. There is no easy answer because people have different values on this.\n\nThe thing that worries me is that these discussions happened a lot when we started the company, before the Ukrainian war. No one—in particular, journalists—was interested in asking us about the capabilities or why we were doing things. Everyone always asked about killer robots, autonomy, our red lines, and whether we had a human in the loop.\n\nWe were always happy to answer. These weren’t annoying questions; they were at the core of what we think about. Now, though, no one asks these questions anymore, and I’m not quite sure why. No journalist ever asks those questions: How are we thinking about the ethical side? We do a lot of this ourselves. We have ethics workshops and a lot of deployment guides for our technologies.\n\nWe have to think very carefully about things like proliferation and a whole bunch of other things, but we have to do it out of an inner discipline. Journalists don’t ask the questions anymore, and the public doesn’t either.\n\nDavid Senra\n\nI don’t even understand the counterarguments that you receive. The argument is, “No, we should not have any new technology.” Humans are violent by nature; we’ve had war since we’ve existed.\n\nTorsten Reil\n\nYeah, but it is an interesting topic because we get this pushback as well, obviously, at Helsing. We get pushback from people who say war is always bad. Of course, at some level that makes sense, but again, if you have a situation where you are attacked, or where there is an adversary that is building up capabilities, you need to be able to deter in order to avoid war. Then you need to be able to fight in order to not lose unnecessary lives on your side.\n\nWe get a lot of pushback, and it is often from well-meaning people. It’s not that we hate people who have a problem with our mission. I disagree with the way they look at it. I think if you think it through from first principles, you need to have this kind of deterrence.\n\nBut we don’t dislike people. Very often, they’re people who just want to have peace. It’s just that we want to have peace, too. We just think you have to achieve it by not letting a war happen and not being inviting to a war.\n\nDavid Senra\n\nBut I think there’s something interesting. It goes back to your high conviction, the fact that you’re extremely disagreeable. You’re very comfortable taking a counterposition. You started this company when everybody else said it was very unpopular. Something that I asked Daniel Ek about, and he said, “He’s a German building a defense company in a country whose entire postwar identity was ‘Never again.’”\n\nTorsten Reil\n\nYeah.\n\nDavid Senra\n\nYou’re also willing to cut across your cultural expectations, too.\n\nTorsten Reil\n\nYeah. Yeah. I think that’s still the case in Germany, and we’re set up as a European company. We have big offices in London and Paris and a bunch of other countries as well, but our biggest office is in Munich, in Germany. German society is a pacifist society, and it’s having to face a lot of reality right now, which has been happening in particular over the last couple of years, where I think Germans realize it hasn’t been the end of conflict.\n\nThere are now conflicts, and they’re coming closer, and they might be coming even closer again. It’s very uncomfortable for a country that thought those times were over, but it’s also the country that now has one of the most thriving new defense ecosystems, I would say, in the world.\n\nSo there is now a rethinking of some of those attitudes. It’s also the country that has, in absolute terms, probably the fastest-growing defense budget, going from 52 billion euros last year to, originally, I think, 153 billion in 2029. I think it’s gone up again to over 180 billion. This is a large increase if you think about the fact that there won’t be more soldiers, most likely, so most of the money is going to go toward new equipment spending.\n\n### Mass Manufacturing, Software & the Future of Materiel\n\nDavid Senra\n\nOkay, so we go back to materiel. Your whole thing is that the winner is going to be able to produce more materiel than their adversary. I still don’t think I completely understand the way you think about achieving that goal. Is the hardware that you’re manufacturing supposed to be as simple as possible, and then you’re offsetting all the innovation in software? Is that still the basic thesis?\n\nTorsten Reil\n\nThe idea is that you build hardware that is good enough and build it in such high numbers that you eventually have economies of scale, but also the ability to overcome any adversary through pure mass. The way you achieve that is that you need to be able to mass-produce. You need to design to mass-produce.\n\nBut crucially, you need software to allow those systems to, first of all, be smart and be able to coordinate. Ideally, you also absorb a lot of the complexity of the hardware that usually will be these exquisite systems that legacy primes will make into your software. So you can make the hardware simpler. That means your sensors are simpler, and you regain a lot of the accuracy through very good AI algorithms.\n\nIt might also be that your manufacturing tolerances are a little bit bigger, and you regain a lot of the control and accuracy again through software that might recalibrate the system as it goes. This is what we do, and that works extremely well.\n\nBut you have to think it through as one unified system. Once you do that, you can go into mass manufacturing. The HX-2 is now a mass-manufactured strike drone, and we had to go through all of those learnings in July and August last year, when we had to go from dozens of HX-2s per month to a baseline of 1,000. It can now scale to 3,000 a month in the small factory that we have, and it's a completely different kettle of fish from producing just a few dozen.\n\nYou have to think about your supply chain completely differently. Obviously, the way you design the drone, quality control, and tracking every single step—you have to learn this. But once you've done this and you're able to produce in very large volumes, you have much more powerful systems because you can now overcome any adversary or target by having saturation attacks, for example, by sending more and more of these drones, by having them highly coordinated, and, if necessary, just producing more and more of them.\n\nDrones are just one example. The same is true for fighter jets. We want to build many hundreds of these fighter jets. The same is true for other systems, maritime systems. It's going to be about being able to build these things in large numbers, designing them so that they can be built in such numbers, and having the industrial capacity and supply chain to produce a lot.\n\nDavid Senra\n\nHold on, though. If we're talking about war materials, you can never predict the potential new adversary you might have in the future. That means you guys have to be completely vertically integrated. If I'm a country—or a continent, let's say we're European—it means I can't outsource some of this production to China, America, South America, or anywhere else. Correct?\n\nTorsten Reil\n\nI wouldn't say any—I mean, with China, yes, that's true. There are still allies that you can work with, obviously, and that we all work with. But there will be some areas, and it depends on the sensitivities of particular countries, where you want to be fully vertically integrated in that country and have your entire supply chain in that country as well.\n\nThere are other areas where you work with allies, and I think that will continue and probably even grow. Then there are other areas where you say, “No, actually, there are so many different suppliers that it doesn't really matter. We are kind of free to choose.”\n\nDavid Senra\n\nBut I feel like you would go more toward complete control. I'm not talking about what you're doing now, but in the future you're going to be running this company for the rest of your life. You just said, “I'm not selling. This is my last company. This is my grand ambition.” So, would you not want to work toward an area where you have complete control over everything?\n\nTorsten Reil\n\nYes. But you also want to work with allies. There are many countries that you want to work together with, and where you are stronger if you work together. There are some areas where specific countries have sensitivities around particular data. For example, the UK is very specific about what they're doing in maritime, understandably so, where you want to be really just national.\n\nBut then there are other areas where you work with allies, and you need to be able to dial it up and down. So what we're doing is, we have this concept of a resilience factory where, if a country wants to buy and build our hardware, they can do it on their own territory. We put a factory there, and they can also build up their own supply chain to the point where they feel comfortable.\n\nDavid Senra\n\nExplain how that partnership would work.\n\nTorsten Reil\n\nYour technology is manufactured in their factories, on their soil. We can run the factory, but we make sure it's on their soil. We find a way to work together with them to also make sure that the supply chain is as national, as domestic, as it needs to be. Again, the sensitivities in different countries are different.\n\nSome countries don't care at all. They just buy it from us, and we can produce wherever. Others really care that it's built on their soil so that they have more control over it.\n\nDavid Senra\n\nWhy do you call them resilience factories?\n\nTorsten Reil\n\nBecause we designed them not to be one big factory, which is a very easy target for attacks, but to be distributed all over a continent or even underground, as is the case with us, so that they're harder to hit.\n\nDavid Senra\n\nYou studied the Krupp family?\n\nTorsten Reil\n\nNo, I know of them, but I don't know them in detail.\n\nDavid Senra\n\nThere's a great book I'm going to send you. It's about 700 pages. I should do an episode on it for my other podcast, Founders. They ran into this problem during World War II, where the Allies found their centralized factories and bombed the shit out of them.\n\nTorsten Reil\n\nYeah. By the way, this problem is completely underestimated right now. Coming back to fighter jets again, we have these exquisite fighter jets, like the Eurofighter, Rafale, and a bunch of others, but they all need airfields, with the exception of the Gripen. They tend to use air bases to take off and land.\n\nThose air bases are obvious targets for an attack. If you look at something like the CA-1, our autonomous fighter jet, it's designed in such a way that it can take off from any kind of makeshift airstrip, which could be a road, an autobahn, a motorway, a highway, et cetera.\n\n### Airbase Vulnerability, Procurement & Scaling Strike Drones\n\nDavid Senra\n\nEssentially adapting it to where you can't bomb—you're not going to bomb all the roads.\n\nTorsten Reil\n\nYou can't bomb all of it.\n\nDavid Senra\n\nExactly.\n\nTorsten Reil\n\nYeah, and I think certain governments and militaries are still in denial about what is actually going to happen in a conflict. In many cases, we won't be able to take off with our crewed fighter jets anymore.\n\nDavid Senra\n\nSo, how do you go about trying to explain that to them? How do you just make it obvious? Do you do it through product demonstrations?\n\nTorsten Reil\n\nYeah, it's like with everything else: you do product demonstrations and show what the capabilities are. We've done this a lot with drones, and, to everyone's credit, people do get it. It's not like people don't complain that procurement is so slow and governments are so slow. Yes, but then, when they do get it, it goes very fast.\n\nThe large strike drone contracts happened really quickly, with political support and support from the procurement agency. They went through in record time, and now, if anything, I think the companies are really under pressure to deliver all of this stuff very fast.\n\nWe all complained about it in the past, but it's a bit of a legacy problem. I think procurement has gotten faster, and the realization that these things have to happen is happening faster too. There will be big autonomous programs in Europe as well now. 2 years ago there was nothing; even 1 year ago, there was nothing. But it's now happening.\n\nDavid Senra\n\nSomething that just came to mind while you were speaking earlier is that you're talking about how, in the AI age, everything is going to happen faster. If you think it's going to happen in 5 years, it's actually 2—maybe 2 years, maybe 2 months. It might even be faster than you suspect. AI is making your products better and better, faster than you expected.\n\nCan AI also help with manufacturing and increase your manufacturing speed? If you're saying you were going to produce a couple thousand a month, isn't China doing something like 100 million drones a year?\n\nTorsten Reil\n\nNot really. Ukraine is building, I think, 2 million or so a year. But those are different types of drones. They're FPVs; they're quadcopters, and, again, they have their role and they work, but they're much shorter-range and they've got smaller payloads of explosives.\n\nStrike drones are much bigger. I mean, they're like this.\n\nDavid Senra\n\nHow big is your strike drone?\n\nTorsten Reil\n\nThe payload—for example, the explosive—is 2 to 4 kg, or even larger. The range that they can fly is around 100 km. It's quite a lot more than FPVs, and they're used in different ways as well.\n\nThose numbers right now—up to 3,000 a month—are really quite high for this kind of category, and you can scale it up even further. At the moment, honestly, the bottleneck isn't manufacturing speed. We can build them really fast. There will be a point, though, in the future where you want to be even more efficient and faster, and then AI will play a role in manufacturing as well. At the moment, it's too early because we're not limited at the moment.\n\nDavid Senra\n\nSo wait, what is the bottleneck then?\n\nTorsten Reil\n\nHonestly, the bottleneck right now is that the big programs need to come. The first ones are coming. We just closed a €1 billion framework contract in Germany, but they now need to come everywhere, and then we can crank up manufacturing accordingly.\n\nDavid Senra\n\nWhat do you mean by everywhere?\n\nTorsten Reil\n\nIn the various countries.\n\nDavid Senra\n\nIn all the Western countries?\n\nTorsten Reil\n\nYeah.\n\nDavid Senra\n\nOkay.\n\nTorsten Reil\n\nYeah. There'll be some in Asia as well.\n\n### Talent Density, Hiring & Performance Management\n\nDavid Senra\n\nAll right. I want to get into how you think about running the company. You mentioned earlier that the mistake you made with your previous company was not putting enough emphasis on talent and recruiting. What did you mean by that?\n\nTorsten Reil\n\nWhen you build a company, you obviously try to hire good people, and every founder will always say, “The most important thing is talent and people,” and they spend so much time on recruitment. But when you actually look at it in detail—and I include us at NaturalMotion previously as well—it is, I wouldn’t say half-assed, but it isn’t the most important thing.\n\nI would argue that hiring and finding the right individuals, number 1, and then, secondly, curating that group of individuals and also moving people out, is the most important thing in a company. It is the reason why a company has great culture, and it’s also the reason why a company is high-performing and feels light-footed. Helsing feels that way.\n\nI think it feels like a very fast, highly performing company that can adapt really quickly, and it is almost entirely because the talent density is so high. So if that’s your thesis—and it was our thesis—then you have to think about how to solve it, and there are 2 main factors.\n\nThe first one is that you need to hire well. You need to build a machine that hires well and be able to monitor how it does it. The second one is that it needs to be able to curate as well, so you need to have a machine, a system, a set of processes that ideally are very easy and allow people to also manage people out when they don’t work out.\n\nWhat a lot of people don’t appreciate is that these 2 are linked because it becomes much easier to move people out who are not a good fit for the company if you know your hiring pipeline is so good that the position will be filled almost immediately. Then it becomes a low-opportunity-cost thing.\n\nWhat happens instead in a lot of startups, and probably also larger companies, is that managers are hesitant to move people out of a company because, A, it’s always unpleasant, so you need to hold them accountable, but, B, because they may rightly know that it will take 9 months to fill that position again.\n\nVery often, a manager’s argument will be, “Yeah, but if I fire this person now, then we won’t be able to hit our deadline because at least they’re going to get the job done somehow. Maybe a bit slower than they should, maybe they create cultural problems, maybe they have an attitude issue, but at least eventually they’ll get the job done.” That’s not good enough. You need to always have enough pressure on the pipeline to be able to fill those positions quickly so that it becomes easier.\n\nWhat we did with Helsing at the beginning, because we knew this was going to be important, was sign our first contract—even before we founded the company—with a third-party recruitment agency because we didn’t have anything internal. We always wanted to start with pressure on the recruitment pipeline.\n\nGundbert, Niklas, and I then spent easily the first year reviewing the KPIs every week to see how our recruitment worked, and we always injected ourselves if it needed to be readjusted because the recruitment pipeline was so important. Building the internal recruitment team was really important, and then building processes for how you manage performance was important as well.\n\nThe most important thing is a quarterly performance-calibration review. What we do there is we don’t actually review the performance of all the individuals. We listen to the managers explain why they give their reports a certain rating, and then the manager is held accountable to that rating.\n\nWe ask them, “Why did you give that person that kind of rating?” Then we might say, “Well, that doesn’t sound like a 3. That sounds like a 2,” or, “That doesn’t sound like a 3. It sounds like a 4.” That allows us to make sure that people have the same calibration of what good performance is and what good looks like across the company.\n\nIt also means that it’s hard to hide if someone is an underperformer. If you give them a low rating, the expectation is that you do something about it. You either make sure that they become better—that’s your task as a manager—or, if that’s not possible, you find ways to part ways.\n\nThis kind of accountability is really useful for a company because it allows you to eventually scale up and have a general view of what the benchmark is that we’re looking for. It also gives us a feeling for which managers try to protect their teams and which don’t.\n\nDavid Senra\n\nI’ve got to pull something up. Do you remember we were together—we were racing cars in Sweden, and Luca Ferrari was there?\n\nTorsten Reil\n\nYeah.\n\nDavid Senra\n\nDid you spend any time with him?\n\nTorsten Reil\n\nI didn’t know him.\n\nDavid Senra\n\nYou need to, or I’m not telling you what to do. You should. How about this?\n\nWe had this crazy lunch, and Luca is a very—I like him a lot—but he’s a very strange cat, in the sense that I consider the way he runs a company almost like a Galápagos Island of entrepreneurship. He didn’t look to other entrepreneurs. He doesn’t really know much about entrepreneurial history when I talked to him. He doesn’t really pay attention to anything. He’s just essentially trying to figure out how to run a company based on what he wanted to do.\n\nSo he’s got a lot of very unique ideas. But I remember we had these notes, and we were just going back and forth on WhatsApp. I was like, “Dude, I’m just going to type all the notes of the crazy shit that you said during this lunch because you have some very interesting ideas on talent and specifically HR.”\n\nHe had this idea. These are some notes from this. You should spend some time with him. He talks about hiring. He says, “We make it a science.” He’s talking about what he’s doing at Bending Spoons. He has a team of 50 in HR, but they’re all engineers.\n\nI’ve never heard of this before. They compare 1-, 2-, and 3-year performance to signals from their CV. He’s saying, “Hey, we think quality X is really important, but we’re just guessing. So what we’re going to do is say, ‘We hired this person because of quality X. Well, what’s their performance based on all the people we hire for quality X, 12 months out, 24 months out, 36 months out?’”\n\nThey’ve actually developed an algorithm of 100 different signals that they figure are important for talent. He’s like, “We treat it as an engineering process.” Then he says—remember, he has all these different companies—“We centralize hiring and firing.”\n\nHe says, “Talent managers are in charge of both entries”—people coming into your company, which you just described—and then exits, which you also just described. His whole thing was that you should have brilliant people in HR. Most companies treat HR as B, C, and D players. He’s like, “No, you should have the best people in HR because it’s so important.”\n\nTorsten Reil\n\nTotally agree. I think this kind of investment, taking it so seriously, just compounds over time because you get the best people in. It carries a different culture, and it becomes something that will be central to being successful eventually, especially if you can keep that muscle over time as you grow as a company.\n\nIt’s easy to lose it over time, and we always say a company like Helsing has to shed its skin almost every 9 months. You have to rethink how you do your processes. You have to rethink how you grow and what kind of leadership you need. But if you can keep that kind of investment in making sure you have a great intake of people, it’s super valuable.\n\nDavid Senra\n\nAre one of the co-founders having to approve every hire? How do you guys do this?\n\nTorsten Reil\n\nWe don’t have to approve every hire. I think that’s too difficult now. I think we do get involved in hiring.\n\nDavid Senra\n\nI’m going to argue with you since you like to argue. You have how many employees?\n\nTorsten Reil\n\n1,300.\n\nDavid Senra\n\nOkay. It’s not too difficult. Elon personally approved the first 3,000 hires of SpaceX. That’s in one of the biographies.\n\nTorsten Reil\n\nApparently. Yeah. If you know, that’s great. We don’t do it that way. We have very good leadership that is able to make those decisions.\n\nIt is true that sometimes we get pulled in for a second opinion, and it’s also true that we get pulled in to interview people and sometimes to convince them to come into the company. Then, for certain roles, we do a lot of the hiring ourselves. We—the founders and executive team—spend a lot of time on hiring.\n\nDavid Senra\n\nI want to hear more about this.\n\nTorsten Reil\n\nFor key positions, we will sometimes do the hiring end to end as founders. We’ll do the search, work together with search agencies, and work together with our network because those positions are so important that we want to get it right. We know that if we get the right person, it’ll be transformational for the company.\n\nDaniel also gets involved in our hiring. He played a very big role in hiring one of our most senior executives recently, which went really well. So we all chip in. It’s so important to get the right people in. It is, again, the most important thing.\n\nDavid Senra\n\nBut if somebody’s in front of you, how do you actually know if they’re good or not?\n\nTorsten Reil\n\nEveryone, I think, has different methods. For me, I try to talk to them and get a feeling for the person without a standard set of questions. I just try to find out whether a person is effortless. That’s probably the most important characteristic I look for.\n\nDavid Senra\n\nWait, what does “effortless” mean?\n\nTorsten Reil\n\nIt’s like one of those things you can tell. You can tell if they don’t try to impress you, if you get a feeling that they have deep knowledge of what they’re talking about, and at the same time they’re still curious and have high agency. That’s the combination I look for. That combination, to me, is effortlessness.\n\nI don’t think it takes that long to find out. It takes maybe 10 minutes in a conversation to get a good feeling for whether a person has this kind of effortlessness or not. That’s what I look for. But I don’t interview software engineers, for example. That wouldn’t apply there.\n\nDavid Senra\n\nBut I mean, you’re essentially involved, I would imagine, at the most key positions that you’re trying to recruit for. What do you think is happening in these 10 minutes?\n\nTorsten Reil\n\nHonestly, I haven’t thought about it. I don’t know. I just try to get a feeling for the person. I don’t reflect on these things that much, to be honest. I just know it works for me, and the only thing that matters to me is whether the outcome is right.\n\nI do obviously look at the statistics, because if I came in and just vibed the whole thing and everyone, and it ended up being crap after 6 months, obviously that wouldn’t work. But that’s not the case. All the senior positions are obviously interviewed by all the founders, so we always have a second and third opinion on them.\n\nWe tend to agree on people. I think we look for similar things. Sometimes you have to go into longer interviews, and people do longer interviews with us as well. For me, it works better if I do short interviews and just get a feeling for the person.\n\nDavid Senra\n\nHave there been examples where your initial intuition about a person has been proved wrong, in Helsing in particular?\n\nTorsten Reil\n\nNot at Helsing, but 3 times at NaturalMotion. I still know the names, and it still really annoys me why I didn’t see it early enough. In all 3 cases, they were so good at faking the effortlessness and so good at coming up with fake facts that I didn’t see through it.\n\nDavid Senra\n\nI heard you say something on Harry’s podcast. Did you actually say that you think there are a lot more exceptional people out there than other people believe, especially if you have a couple thousand employees?\n\nTorsten Reil\n\nPeople’s character and performance are not set in stone. Quite often, in larger companies, it might actually be harder for people who have 10× potential to become 10× people. It’s a combination of things. It’s probably harder to keep hiring 10-out-of-10 people, but it’s also harder to develop them.\n\nBut it’s still worth it. The power law in talent is extreme. There are very few 10-out-of-10 people, but they make—I mean, it’s a cliché, but we see this every day—a disproportionate difference to the company. So it is worth trying to find these people, and it is worth finding a way to give them a home in a company where they can really thrive.\n\nWe try to do this very systematically at Helsing. We create, hopefully for everyone, a great environment where they’re surrounded by amazing people. But for the 10-out-of-10 people, we go the extra mile.\n\nDavid Senra\n\nHow do you do it systematically?\n\nTorsten Reil\n\nI don’t want to go into too much detail about how we do it, but we create an environment where they can do their best work. We make exceptions that we wouldn’t always make for everyone because they’re such big contributors to the company, and many of them have contributed so much already.\n\nIt’s also a feeling of gratefulness, to be honest, to have a bunch of those key employees who have made a big difference. You can point at the successes of the company and always know that they’re there. Obviously, it’s all about talent because people make all of this stuff. Maybe in the future it’ll be more AI, but at the moment it’s still mainly people.\n\nYou can point at those people who made a difference to the company. So we all feel a lot of gratefulness to those top individuals—to everyone in the company, obviously, but in particular to the ones who have turned a project from red to green. That happens very often, and it’s quite often very small teams or specific individuals that make a big difference.\n\n### Helsing's Culture: Whatever It Takes & Truth Is in the Field\n\nDavid Senra\n\nBased on the conversations we’ve had, the way I would summarize your perspective on this, because you talk about this a lot, is talent density above everything else.\n\nTorsten Reil\n\nYeah. I think it’s the number one thing. I feel that culture comes from talent density.\n\nWe had a situation—I don’t know, maybe it was 18 or 24 months ago—where we had a set of values at Helsing that we came up with at some point, I think in a workshop or so. They were on our website, and they were fine, I think.\n\nThen we had one of our get-togethers where the whole company got together, and a bunch of people came and said, “We want to rewrite our values. We want to throw the old ones away. We know now what Helsing is about, and we want to do it completely differently.”\n\nAt this get-together, there was a workshop, and I have to say I was dreading it because I didn’t want to do another values session. I never like them. But it was effortless. The team came up with their own values based on their experiences. It wasn’t one of those exercises where everyone contributed ideas. It was just kind of a mind meld, and it was clear that this was what it was going to be.\n\nThe top value was “whatever it takes.” Everyone gave examples of why that was. Some had just come back from the front line in Ukraine. Some had not slept for 48 hours and had done a whole bunch of things that were just amazing and impressive.\n\nAnother value that came out of it was “the truth is in the field.” It doesn’t matter how much you claim that your system works in tests, government tests, and so on. We found out that it has to work in reality and in combat. So “the truth is in the field” came out of that.\n\nI was so impressed by the team, and I was thinking, “How did that part of the culture come about?” As far as I can tell, it came about because a bunch of really good people got together, really cared about the stuff, had initiative, and thought about it. Then they came up with it. It was just that.\n\n### Deploying in Ukraine & the Weight of Defense Technology\n\nI think getting really good people together who have good chemistry—and when they sometimes have the sparks flying, finding a way to work together again—that’s what creates culture.\n\nDavid Senra\n\nSo you have this idea to start this European defense company. You have ideas for the products you want to build, and you raise money. Then you go from the idea to your products actually being used in Ukraine for the first time. Do you remember the first time you found out that your product was being used there? Where were you, and what did that feel like?\n\nTorsten Reil\n\nWe knew from our team in the field. I do remember it, and everyone thinks differently about these things. I respect most attitudes toward it, but I personally found it very difficult.\n\nI think it became a more serious company, and some of the lightheartedness was gone a little bit. If you build a video game company or a SaaS company, you can be quite lighthearted and unserious about things. It’s a commercial business, but you can be that way; it fits culturally.\n\nIf you’re deployed in what is fundamentally a sad thing—this kind of conflict—and you contribute to it because you have to, you can think it through from first principles, but in my opinion it’s still sad that it happens. That makes it a serious thing. It changes the culture, and we’re very aware of it.\n\nThat doesn’t mean that we don’t want to create the best systems, of course, and it doesn’t mean that we’re not super ambitious. We are. But it is serious. It’s very serious. We don’t take it lightly.\n\nDavid Senra\n\nAll right. You’ve used the word “deterrent” multiple times throughout this conversation. So that means the best possible outcome for Helsing is that nothing happens.\n\nTorsten Reil\n\nYeah. You don’t want a conflict. In general, obviously, the reason why you build capabilities is in order to avoid conflict, not to use them in a fight.\n\nDavid Senra\n\nWhich goes back to this dinner we had about 2 years ago, this argument we were having about the fact that war has been constant throughout human history.\n\nWhat actually has to take place, in your mind, for that outcome to happen where there’s no more human conflict?\n\nTorsten Reil\n\nThat’s a big question. I do think it’s in human nature to have at least parts of the population that want to have conflict, that enjoy conflict, or that want to increase their sphere of influence. Sometimes that is political, sometimes it’s geographical, and sometimes you literally just want to have more land around you.\n\nI think that is partly cultural, but it’s also partly hardwired into at least a part of the population and will probably be genetic in a part of the population. In order for that not to dominate everything else, you need to make sure that the rest of the population is sufficiently protected so that it’s too expensive to try and use force to take what you want. That’s basically what it boils down to.\n\nIf you could eliminate the outliers, let’s say, who are really that aggressive and want to use force, then I think forever you’d have peace. It’s unlikely you can eliminate it. I think it’s more likely that you have to contain them, and I think that’s okay as long as you see it early enough that you have to contain them and are able to look very prickly to them so that they don’t attack.\n\n### NaturalMotion, Gaming & Torsten's Path Into AI\n\nDavid Senra\n\nWe mentioned your previous company a few times, and you said something about how, when you guys were developing the autonomous fighter jet, you used a very similar technique to what Demis Hassabis used to do AlphaGo. I just finished reading this biography of Demis. It’s interesting to me that he got started—his first introduction to AI was through video games. Yours was, too. Some of the best entrepreneurs come from gaming. What do you think it is about that?\n\nTorsten Reil\n\nYeah, and I think arguably Palmer Luckey did, too. I mean, it was certainly very close to the gaming industry. Gaming companies are complex to build because you marry art with technology. In many cases, especially with console games back in the day, they were some of the most sophisticated IT projects that you could think of, and at the same time they had to have artistic value as well.\n\nYou had to accommodate people from all walks of life, like very focused software engineers with artists who were just trying to create something beautiful. That is not dissimilar to what happens in the defense industry. Maybe there’s a little bit less artistry, although that exists as well, but we have very different people who work on very complex systems, who have very different cultural backgrounds, and still need to work together to produce something amazing in the end.\n\nI think that’s one part, and maybe it’s also exposure to defense through games. I don’t know. A lot of games have a military theme. Maybe that’s part of it as well.\n\nDavid Senra\n\nExplain the founding of your first company. What was happening right in your life before you started that one?\n\nTorsten Reil\n\nI was studying in the UK, and I was working on biology. I was working on systems and simulations that simulated neural networks.\n\nDavid Senra\n\nIn what year?\n\nTorsten Reil\n\nThe year was 1999.\n\nDavid Senra\n\nWait, you were working on simulating neural networks in 1999?\n\nTorsten Reil\n\nYeah, using genetic algorithms. I was looking to optimize them to do something useful.\n\nDavid Senra\n\nWhy were you interested in that to begin with?\n\nTorsten Reil\n\nI liked complex systems. I had just started a PhD in complex systems in biology, and neural networks are a complex system. Specifically, the ones I looked at were fully recurrent neural networks, the ones that, with the right weights, could start to create spontaneous cyclic activity.\n\nDavid Senra\n\nIs that not what you just described on the fighter jet?\n\nTorsten Reil\n\nWell, it has some similarity because genetic algorithms have fallen out of favor as a search algorithm, and now it’s almost entirely reinforcement learning. But you could theoretically also use genetic algorithms, so there’s definitely some similarity. The network architectures are different.\n\nBack then, I based my network architectures on biological, what are called central pattern generators, or CPGs. They’re the neural structures that create rhythmic movement for walking, your heartbeat, and a bunch of other things. I tried to recreate one of these for walking using genetic algorithms and a neural network—a quite simple neural network—to optimize the weights.\n\nIn order to see whether it worked, I hooked it up to physically simulated creatures: very simple four-legged and then two-legged creatures. Physics engines had just started back then, with MathEngine, Havok, and other companies, and to my surprise, it ended up working quite well.\n\nI remember it’s quite rare that you have these eureka moments, but I had this thing set up overnight, and those computers were very slow back then. The next morning, when I looked at the experiment, I literally had this two-legged creature walking over the screen, fully physically simulated.\n\nI did a bit more research in that area and then realized this was going to work very well. I thought I’d spin out a company to use these things as virtual stuntmen in movie scenes.\n\nDavid Senra\n\nIf I go back to you in 1999, did you have any inclination that one day you’d be a founder or an entrepreneur?\n\nTorsten Reil\n\nNo, not at that point.\n\nDavid Senra\n\nWhat did that person in 1999 think they were going to do?\n\nTorsten Reil\n\nI thought I was going to be a scientist. I wanted to stay in science and be a biologist, and specifically do software and simulation work in biology. That’s what I wanted to do.\n\nDavid Senra\n\nOkay, but then you have this eureka moment, and your first thought is, “I’m going to start a company”?\n\nTorsten Reil\n\nIt was my first thought. It took a little bit of time, but this was a time of the dot-com boom, and everyone wanted to start a company.\n\nDavid Senra\n\n1999 was an important point.\n\nTorsten Reil\n\nYeah, and it still looked like it could do that kind of thing. It was inspiring to see lots of other people start companies. So I started NaturalMotion, and we first used it to produce software called Endorphin, which was used in big Hollywood movies like The Lord of the Rings, Troy, Spider-Man, and Harry Potter.\n\nThat was fun and great, but we didn’t make much money because we only charged $50,000 per license, and they ended up doing all of the scenes with it. It was clear that the real value was going to be in using that technology for games in real time.\n\nRockstar Games was one of our customers for the old tool, Endorphin. They asked us, “Do you have a version that runs in real time on the upcoming consoles?” And we said, “Kind of.” So then we started a long collaboration with Rockstar Games, which ended up using our technology in Grand Theft Auto IV, Grand Theft Auto V, Red Dead Redemption, and a whole range of other games.\n\nThe company grew, and it grew faster than before, but it still didn’t have the kind of takeoff—the kind of liftoff characteristics—that I was hoping for as a founder. You always hope for more of a hockey-stick thing, and it felt like the chess pieces were drawing in and it was going to be hard to move.\n\nSo we ended up saying, “We need to make games ourselves,” and what you would now call vertically integrated. We didn’t think about it that way. We felt like we were so inspired by and had learned so much from Rockstar Games that we thought, “How hard can it be?”\n\nIt turns out to be quite hard, but it ended up working quite well. We acquired a games team and made the first games. The iPhone came out, and then we had a string of hit games on the iPhone, followed by a big breakthrough in a completely different area.\n\nAll of our previous games had been about characters because that was our technology. Then there was one game that I really wanted to make because I was really into cars: a drag-racing game, a car-racing game without any steering.\n\nThat became by far our biggest hit, CSR Racing, and became the biggest game of all time on the iPhone at the time. That gave us the hockey stick. Then it was liftoff for the company, and it became a completely different experience.\n\nDavid Senra\n\nI feel there’s some theme connecting all of this, and I still don’t have an understanding of it. You’ll see this in all these biographies that I read. I’m reading one right now: Raymond Plank writes it when he’s 90 years old, and he starts this company called Apache Corporation with $250,000 in 1954. By the time the book is published, 50 years later, it’s a $30 billion company.\n\nHis insight was that he was a tax planner and accountant, and some of the businesses he was working for were these new American oil companies. He realized there were huge tax incentives taking place in America after World War II. At the same time, there were tax benefits if you invested in oil discoveries because the United States was trying to incentivize domestic oil production.\n\nAnd because there was so much debt after World War II, they increased the top marginal tax rate to around 90 percent. The very wealthy were trying to find ways to offset that. That was the original idea. He discovered Apache Corporation, which wound up being this massive oil and gas company, because of the work he was doing in tax planning and accounting for relatively small businesses.\n\n### Outsiders, First Principles & Ambitious North Stars\n\nI feel there’s some kind of connection between your first company and your second that I’m trying to get to.\n\nAm I just wrong on this?\n\nTorsten Reil\n\nI don't think it was that direct, necessarily. One theme that I've always felt with everything I ended up doing is that I ended up doing something slightly different from what the mainstream was doing. Part of the reason is that I've never felt part of the peer group.\n\nAs a biologist, I did a lot of biology and loved it, but I never really felt like I was part of the biology group. That made it a little bit easier to do different things. That's why I ended up doing software and simulation rather than going out into the field to watch animals and dissect animals. I never liked that stuff.\n\nI wanted to do some slightly different stuff with gaming. I also never felt part of the gaming group and culture in the same way. As a result, I think we ended up doing quite different games. I remember that the pricing for our games was very different from anyone else's.\n\nDavid Senra\n\nThe pricing?\n\nTorsten Reil\n\nYeah. The first games that we made, we priced as low as we possibly could because we felt that we just maximized revenues that way. Everyone else in the game industry said, “No, your game needs to have a value. Your mobile game needs to be at least $5 so that people feel it has a value.”\n\nIt never made sense to me because all I wanted to do was maximize revenues. I felt like there was so much price elasticity that we could just go all the way down and people would spend. Overall, we would make much more money because we'd have many more people. We sold our game for 99 cents, the cheapest that was possible.\n\nThen, when free-to-play came, we were the first big 3D free-to-play game, and we made it free and then charged money elsewhere. That would have been much harder had we been part of the group. I think a similar thing is also true for defense. It would be much harder for us to build Helsing if our backgrounds had already been in defense companies. It would have been just much harder.\n\nIt's easier to be an outsider. That may be a parallel. But the other thing is that I always try to do things that are different and sometimes cool. I really wanted to do the number-one top-grossing game in the world.\n\nI remember standing in the shower at one point in the UK and thinking, “I would like to do the number-one top-grossing game.” It was just a goal. With Helsing, I think we all want to build a company that has impact. I think that's similar to you: You just want to have those goals. You want to have a bit of a north star and then go for it.\n\n### Why Commentary Is Worthless & Founders Should Build\n\nDavid Senra\n\nOkay. I love that you use the word “north star.” The one other thing—I think probably the last thing I want to talk to you about today—is that I've heard you say this criticism of this LinkedIn or X commentary culture: that commentary is actually relatively worthless or completely worthless. What's your opinion on this? Can you just expound on that?\n\nTorsten Reil\n\nYeah, I think commentary is entirely worthless, and it has almost no positive contribution. The only thing that it does is drag some people down. So it has a net negative benefit, and it is obviously encouraged through the way social platforms are incentivizing contributors.\n\nYou're incentivized if you comment, particularly if you comment negatively, but you're not incentivized to be a creator. I don't complain about this because it is what it is, and they should do what they want. But I do think that creators are still somewhat undervalued and commentators are still overvalued because they get so much airtime, and there are many of them now—perhaps even more than before.\n\nIt contributed to some of the European perceived malaise: people talking themselves and each other down. I just said this the other day again. I was in a car with someone who is actually a great entrepreneur, but he himself just kept commenting on how bad certain things are.\n\nThere's no point in commenting about it. It creates zero value. You either do something about it or you don't talk about it. But if you don't do either, then I don't think it's a good contribution. Just commenting on negative stuff is not useful.\n\nDavid Senra\n\nWhat would your advice be to do then?\n\nTorsten Reil\n\nJust focus on building your company. I always think that you have a real privilege as a founder: You can make a difference. Depending on which sector you're in, you can make big differences in the way a society works.\n\nArguably, all the social networks have made a massive difference, for better or for worse. All the AI companies are making a massive difference, as are many health companies. I think defense companies are making a big difference too. If you are a founder who is very ambitious and you're in one of those fields, you can change society, and not many categories of people can do that.\n\nPoliticians can as well. Certain types of other people can as well. Sometimes even investors, but founders, I think, have the biggest lever. That's also why you have so few founders, I think, on Twitter and other social platforms: It's already very satisfying to do the stuff that you do.\n\nYou don't then need to comment on how bad everything is. You just do your stuff, and it has an impact.\n\nDavid Senra\n\nThat's the perfect place to end. Thanks for making the time.\n\nTorsten Reil\n\nCool. Thank you.","tldr_md_alt":"- **Reil的核心判断是：坦克和火炮时代已经结束，边境部署10万架无需GPS、能够自主搜寻目标的攻击无人机，就能阻止欧洲发生陆战。** 一名俄罗斯士兵在前线的生存时间据称“约20到30分钟”，不过Reil称这只是大致估计；如果3000辆入侵坦克面对10万架无人机，“我认为几个小时内就会被摧毁”。一年前他因这一判断受到抨击——“我现在仍然这么认为。”\n- **干扰——无人机主导论最常见的反对理由——正越来越多地被绕开。** FPV无人机可以使用光纤电缆，大型无人机则能借助AI在无GPS环境下导航，算法“白天、夜间、下雪时”都能工作，即便地形已被炮火打得面目全非；战争因此变成“机器对机器”，成为一场物资消耗战，而派人上场既是自杀，也是Reil所说的“不道德”。\n- **Helsing的AI飞行员先是在模拟环境中击败人类飞行员，随后在真实战斗机测试中取得了完整控制权，削弱了有人战斗机存在的理由。** 这套系统以AlphaGo式强化学习训练，运行速度为实时的20–25,000倍；去年它在波罗的海上空完全接管一架Saab Gripen，并与人类控制的Gripen进行了3种空战场景对抗——人类直觉这座“最后的堡垒”“结果证明是胡扯”。随后Helsing收购了慕尼黑以外的一家飞机制造商，CA-1 Europa将于明年首飞，设计上可从高速公路起飞，因为空军基地会成为攻击目标。\n- **制造业路线很明确：硬件做到“够用”，把复杂度吸收到软件里，然后靠纯粹的规模取胜。** HX-2攻击无人机的产量已从每月几十架提升到1000架的基准，目前可以扩至3000架；瓶颈是合同而不是工厂——德国一份10亿欧元框架协议刚刚落地。Reil称德国去年的防务预算为520亿欧元，原本预计2029年达到1530亿欧元，如今可能已超过1800亿欧元；由于大概率不会增加士兵数量，绝大多数资金将流向装备。\n- **Helsing成立于2021年3月，当时银行把防务与“色情、赌博和吸烟”归入同一类别——公司身边有些人因此被银行拒绝服务，没有一家VC愿意接触，于是800万欧元种子轮来自个人和家族办公室，之后Daniel Ek又投入了1亿欧元自有资金（“理想情况是1亿欧元”，这是Reil估出来的数字）。** 这笔资金支持了“正式项目”战略：Helsing与Saab合作拿下Eurofighter电子战升级项目，成立2年4个月后从黑马胜出；Reil认为，这可能是公司从成立到进入正式项目最快的一次。\n- **关于退出，Helsing的立场绝对明确：“我们永远不会出售公司。”** 成立几个月后，创始人决定“全油门冲刺，要么彻底撞毁，要么取得巨大成功”——Reil认为，欧洲真正缺的不是监管或税收，而是创始人的进取性：“一家公司的成功有95%……其他都是借口。”\n- **在运营层面，人才密度高于一切——Helsing成立前签下的第一份合同，就是与一家招聘机构签订的。** 持续的候选人管线压力让解雇成为一件“机会成本很低的事”；季度校准评审让管理者必须为自己的评级负责；Reil用10分钟阅读来判断高级人才是否具备“毫不费力”的感觉——深厚知识、好奇心和高主动性兼具。他回忆Helsing没有出现过这种误判，但NaturalMotion曾有3次。\n- **对Helsing而言，最好的结果是什么都不发生：通过改变“对方的成本计算”实现威慑。** Reil不安的是，乌克兰战争前，记者还会反复追问杀手机器人、红线和人类是否必须保留在决策链中；如今“已经没人再问这些问题”，机器决定生死的伦理只能依靠公司自己的“内在自律”。","digest_md_alt":"### 1. 坦克时代结束——前线生存以分钟计\n- Reil开场就说，在乌克兰，“坦克基本已经不再使用”，火炮也是如此。一名俄罗斯士兵在前线的生存时间据称“约20到30分钟”——他承认这只是大致估计；而待在坦克里更危险，因为坦克是最高优先级目标，很容易被ISR无人机发现，再由小型或大型攻击无人机迅速摧毁。“我认为坦克时代大概已经结束了。”\n- 即便藏在前线后方40–50公里处的火炮，也会被无人机发现并打击。Senra追问分类后，Reil解释称，FPV无人机是小型、主要靠人工操控的四旋翼机；像Helsing的HX-2这样的攻击无人机则能飞行很远，而且“经常由AI控制”。\n\n### 2. 10万架攻击无人机足以威慑欧洲陆战\n- 那条让Reil“惹上了一点麻烦”的推文，他现在仍然坚持：如果3000辆坦克面对10万架无需GPS导航、专门猎杀坦克的攻击无人机，“几个小时内就会被摧毁”。Senra反过来问：如果一个国家有这样的防御体系，还怎么入侵？答案是“非常困难”：入侵方可能会避开这类有人载具，转而使用更小型的自主系统——“基本上一切都在朝这个方向发展”，包括无人机、自主战斗机和地面机器人。\n- 为什么自主能力不可妥协：乌克兰数千名熟练操作员的规模，“西方做不到，NATO也做不到”。人类仍会留在决策环或控制环中，AI负责飞行。当前进展“基本已经解决”：正在建立的直接能力包括末端制导，也就是最后约2公里，以及协同软件；整个领域正向接近完全自主推进。\n\n### 3. 干扰越来越难奏效，战场变成机器对机器\n- 干扰主要有两类问题：回传给操作员的视频链路会被干扰，因此无人机必须自主飞行，或者使用电缆——部分FPV无人机正在采用后者；大型无人机则可以自行飞行。GPS在“几乎整条前线”都受到干扰，于是AI可以像飞行员一样通过识别周边环境来导航。如今算法已经“非常稳健……白天、夜间、下雪时”都能工作，即便地形被炮击炸得凹凸不平。\n- 这也解决了Reil多年前与Senra围绕《Freedom's Forge》式“靠产量压过敌人”论点的晚餐争论：“这将是一场物资之战，是机器对机器。”进一步推演，如果明知道人类上场无法生存，“把人送上战场将是不道德的”；但领土仍然必须夺回并守住，“所以地面需要机器人”。\n\n### 4. 2021年创办Helsing：当防务被与色情归为一类\n- Reil原本没打算在NaturalMotion之后再创办第二家公司，当时他“已经没剩多少能量”；而且也不会独自创办防务公司，因为防务创始人要同时处理技术、政治、出口法、机密信息等复杂问题。他在柏林一家咖啡馆认识了Gundbert，后者当时负责McKinsey的航空航天与防务业务；两人30分钟内做出决定，随后拉上Niklas——他的创业公司为传统主承包商解决AI问题——3人几天内就决定创办Helsing。公司成立于2021年3月。\n- 当时的环境是“非常不受欢迎”。公司身边有些人那段时间被银行拒绝服务；Reil本人没有遇到这种情况，但另一位与公司关系密切的人因为涉足防务而很难获得信贷。银行把保卫一个国家或民主社会“与色情、赌博和吸烟归入同一类别”。没有一家VC愿意接触他们——“最后我们甚至没有尝试”——因此800万欧元种子轮来自个人和家族办公室。\n- 2014年的克里米亚事件是触发点——“我们都不愿相信那是真的”——另一个现实则是，美国有Anduril和Palantir，而“欧洲什么都没有”。“当时的感觉是，如果我们不创办这家公司，就不会再有人来做。”\n\n### 5. Daniel Ek的1亿欧元与永不出售的决定\n- 种子轮完成3个月后，Ek问他们需要多少钱。“理想情况是1亿欧元”——这个数字没有预算依据，“感觉差不多对……几乎所有融资轮都是这样”。Ek通过Prima Materia投入自己的钱，时间早于入侵事件；Reil称他“更像联合创始人，而不是投资人”。\n- 对于退出，Reil的答案绝对明确：“我们永远不会出售公司。”这个决定在成立几个月后做出：要“全油门冲刺，要么彻底撞毁，要么取得巨大成功”，中间没有其他结果。Senra还讲了AppLovin的Adam的故事：股价对应的市值从约280亿美元跌到约30亿美元，借入约60亿美元进行回购，最终赚到约600亿美元——“我不会经营一家30亿美元的公司。”Reil说：“我认为只有这样才能把公司做成。”\n- Senra反对出售公司的立场是：“被收购了。听到这个消息我很遗憾。”Reil则提出他的托斯卡纳理论：欧洲退休生活太惬意了，一旦有几个朋友这么做，社会就会接受这种选择——“这不应该成为社会认可的事情。”\n\n### 6. 欧洲缺的不是别的，是进取性——其余都是借口\n- 使命先于信念：“守护我们的民主”不需要通过一次离地的团建来起草；Spotify则证明了欧洲公司也可以产生全球影响。对于把问题归咎于监管的人，他的回应是：没错，德国GmbH需要公证人——“谁在乎？”防务合规的要求“可能是创业公司要面对的100倍……但它仍然不会拖慢我们。你只需要把公司一开始就搭对。”\n- Senra指出，Reil最令人难以相处的判断，可能就是雄心和进取性占“一家公司成功的95%”。解决方法是树立榜样：“只要有几家极具进取性的公司指出，其他人基本都在敷衍了事。我认为很多人确实如此。”\n\n### 7. 从黑马到正式项目主承包商：28个月\n- Reil认为这个想法来自Gundbert，并由1亿欧元资金解锁：直接瞄准正式项目，也就是规模大、已有预算的政府项目，理想情况下还能与主承包商建立直接关系。初创公司通常避开这类项目，因为资质认证耗时太长，远超它们验证产品、获得证明的节奏。有了1亿欧元，“如果前两三个项目不顺利……我们手里仍然有足够的钱”。\n- 第一个胜仗是德国Eurofighter电子战升级项目。许多防空雷达已经变成软件定义系统，“基本上是在变形”，旧的探测方式因此失效；Helsing用深度学习恢复了解读这些雷达的能力，并与Saab合作击败规模更大的联合体，成为“黑马”，距公司成立仅2年4个月。Reil认为，Helsing可能是从成立到进入正式项目最快的公司。\n- 此后这成为公司一贯的模板：“我们试图寻找规模足够大的正式项目，并判断自己能否以差异化方式贡献一些东西。这基本就是我们的视角。”\n\n### 8. 飞行员说不可能存在的超人级战斗机飞行员\n- 2年前，Helsing给自己定下目标：为超视距空战打造一名超人级AI飞行员。首席科学家Antoine Bordes从Meta加入，此前负责或共同负责Facebook AI Research。空战“本质上是天空中的国际象棋”，通常是2对2，涉及雷达、导弹、能量管理和重新接战，“看起来几乎像芭蕾”。现役飞行员告诉Helsing，这是空战的最后堡垒，人类直觉不可能被替代。“结果证明这是胡扯……AI强得多、强得多。”\n- 原因在于AlphaGo、AlphaStar式的强化学习：系统在速度为实时20–25,000倍的模拟器中运行，探索的搜索空间是受飞行学校训练的人类无法触及的。Helsing会定期让它与现役Eurofighter、Rafale和Gripen飞行员对战。\n- 证明能力的关键节点是：AI先在自有模拟环境中获胜，随后又在一家大型主承包商提供的官方模拟器中接受测试；去年，它在波罗的海上空“完全接管”一架Saab Gripen，与人类控制的Gripen进行了3种空战场景对抗。那一刻，Helsing意识到“有更大的事情正在发生”。这套名为Centaur的系统如今还能在战斗机和无人机组成的蜂群中运行，而且“扩展性非常好”。\n- 这也说明了时间表正在缩短：10年前，所有事情都被认为“还有5年”，最后往往耗时7–10年；现在，“当你认为某件事还有5年时，它实际上通常只需要2年”。\n\n### 9. CA-1 Europa：买下飞机厂，从高速公路起飞\n- 如果与一家主承包商合作开发自主喷气机，“耗时会太久”，因此Helsing收购了慕尼黑以外的一家飞机制造商，连同跑道、机库和完整制造能力，围绕Centaur打造CA-1 Europa。飞机计划于明年首飞，目标产量为“数百架”，并希望“在这一类别中跨越所有竞争者”。\n- 对于Senra在Michael Dell家中调侃的监管问题，Reil解释称，原型机前排会安排一名安全飞行员，因为一旦坠毁，落点是“人口密集的巴伐利亚”，而不是莫哈韦沙漠的沙地；这种安排不需要改变设计，就像Waymo早期配备安全驾驶员一样。\n- 一个被低估的脆弱点是：Eurofighter和Rafale这类高性能战机都需要空军基地，Gripen除外，而基地是显而易见的攻击目标；CA-1则可以从任何临时跑道、高速公路或公路起飞，“你不可能把所有地方都炸掉”。Senra提到二战时期Krupp集中式工厂被发现并轰炸的类比；Reil的判断是，“一些政府和军队仍然处于否认状态”，但一旦演示证明能力，采购就可能“以创纪录的速度”推进。\n\n### 10. 够用的硬件、聪明的软件，以及不该被炸掉的工厂\n- 统一系统的路线是：把硬件设计成适合大规模生产的形态，把复杂度推给软件；传感器精度可以没那么高，再由AI恢复；制造公差可以放宽，再用软件重新校准；最终依靠“纯粹的规模”和饱和式攻击取胜。\n- HX-2的产量在去年7月至8月间从每月几十架提升到1000架的基准，现在可以扩至3000架，而且是在一座小型工厂内完成——“完全是另一个量级”。Senra问中国是否每年生产约1亿架无人机；Reil说“并不是”，转而提到乌克兰生产的“约200万架”另一类无人机，即短程FPV无人机。攻击无人机的载荷是2–4公斤炸药，也可以更大，航程约100公里。\n- 瓶颈不在制造速度，而在项目。一份10亿欧元的德国框架合同刚刚落地，“接下来它们需要出现在各个地方”——包括西方国家和部分亚洲国家。Reil称德国去年的防务预算为520亿欧元；他原本认为2029年会达到1530亿欧元，如今可能已超过1800亿欧元。由于大概率不会增加士兵数量，他预计大部分资金将流向装备。\n- “韧性工厂”提供了主权能力：Helsing会在客户所在国的土地上生产硬件，根据每个国家的敏感性本地化供应链——排除中国，欢迎盟友——并把工厂分散到整个大陆，或建在地下，“让它们更难被击中”。\n\n### 11. 威慑伦理——以及如今没人再问的杀手机器人问题\n- 道德框架是把人员损失降到最低，理想情况是降至0，这也意味着根本不发生冲突；因此需要威慑，即形成“可信的姿态，让对方计算成本后认为不值得发动攻击，因为代价对他们来说会过于惨重”。对Helsing而言，最好的结果确实是什么都不发生。\n- 如果机器比人类更准确，是否应该让机器决定生死？Reil认为这是一个合理问题，但“没有简单答案”。他担心的是，已经没人再提问：乌克兰战争前，记者不断追问杀手机器人、红线以及人类是否留在决策环中——“这些问题并不烦人”；如今“再也没有记者问这些问题”。Helsing仍会出于“内在自律”开展伦理工作坊，并制定部署指南。\n- 对于包括德国在内的和平主义反对意见——Senra借Ek的话说，德国的“整个战后身份就是永不再来”——Reil的回应是：“我们也想要和平，只是认为必须通过阻止战争发生来实现和平。”至于人性，他认为冲突欲望部分来自文化，部分是“硬编码”的，可能还有一部分人带有遗传倾向；大概无法消除这些异常者，只能加以约束，并“表现得非常棘手”。\n- 得知Helsing的无人机已经部署到乌克兰后，Reil说：“我个人觉得这非常困难……公司变得更严肃了。”游戏和SaaS业务带来的轻松感已经消失。“我们不会把它当成一件轻松的事。”\n\n### 12. 人才密度高于一切：招聘机器与解雇算术\n- Reil承认，自己在NaturalMotion犯过的错误，是没有足够重视招聘和绩效管理；这后来成为Helsing的创始设计：甚至在公司注册前，签下的第一份合同就是与招聘机构签订的，3位创始人在第一年每周审查招聘KPI。\n- 承载整个体系的洞见是：招聘和解雇相互关联。管理者会留下不合适的人，因为重新招人要花9个月——“至少他们最终会把工作做完……但这还不够好”。持续的候选人管线让把人调离岗位成为“机会成本很低的事”；季度校准评审则要求管理者为自己的评级负责——“这听起来不像3分，更像2分”——从而暴露出谁在保护表现不佳的团队。\n- Reil面试时不问标准问题，通常只花约10分钟阅读材料，寻找“毫不费力”的感觉：不是试图给人留下印象，而是拥有深厚知识，“仍然保持好奇，并且有很强的主动性”。他回忆Helsing没有出现过这种误判，但NaturalMotion曾有3次——“我至今还记得他们的名字”——有人靠编造事实伪装自己。Senra则反击称，Elon亲自批准了SpaceX最初的3000名员工；Bending Spoons有50名HR工程师，把简历信号与多年后的绩效联系起来。Reil同意其中的原则：“它会随着时间复利。”\n- 文化是人才密度的下游结果：一次公司聚会中，团队自发重写了Helsing的价值观——“不惜一切代价”（有人刚从乌克兰前线回来，连续48小时不睡觉）以及“真相在战场上”，因为系统必须在战斗中有效，而不是只通过政府测试。人才的幂律效应“极其明显”：对于10分满分的顶尖人才，Helsing会破例，因为他们把项目“从红灯变成绿灯”，公司对此负有“感激”。\n\n### 13. 从1999年会走路的神经网络到CSR Racing：局外人的一条主线\n- Reil的起点是1999年攻读复杂系统博士时的一个顿悟：他用完全递归神经网络和基于生物中枢模式发生器的遗传算法，配合MathEngine、Havok等早期物理引擎建模；一夜之间，“屏幕上真的出现了一个两条腿的生物在行走，而且完全经过物理模拟”。这后来变成NaturalMotion：Endorphin以5万美元授权费进入《指环王》《特洛伊》《蜘蛛侠》和《哈利·波特》；其实时技术还被用于Rockstar的GTA IV、GTA V、Red Dead Redemption等作品；随后公司亲自开发游戏，CSR Racing一度成为“当时iPhone上最大的游戏”，最终带来爆发式增长。\n- Senra试图找出他两家公司之间的共同主题，Reil的答案是局外人身份：他从未觉得自己属于生物学或游戏行业的同行圈，这反而让他更容易采取逆向思维——行业坚持5美元定价时，他们用99美分定价游戏以最大化收入；在3D免费游戏上，他们也选择率先押注大规模市场。“如果我们的背景都来自防务公司，打造Helsing会难得多。做一个局外人更容易。”\n- 他最后谈到LinkedIn和X上的文化：“评论完全没有价值……净收益为负”，因为平台奖励的是负面评论者，而不是创造者。他的建议是：“要么做点什么，要么就别谈它。”创始人拥有改变社会的“最大杠杆”，这也是为什么真正的创始人很少在网上发言。","shows":{"name":"David Senra","hosts":"David Senra"}}