0:00:04 You’re keeping people alive, you’re not keeping them healthy.
0:00:10 The view that medicine starts with disease is part of the problem because it should start with health.
0:00:16 One in five, one in six Brits are meant to live to 100 according to the government. How do you finance that?
0:00:19 Is your 401(k) set for you to live to 100?
0:00:25 People age really, really adversely and you can be 100 and literally run in 100 meters.
0:00:28 You can look at it on YouTube or you can be 50 and in a wheelchair.
0:00:34 And instead of this being a crisis, it could be the greatest accomplishment of humanity.
0:00:37 Let’s start this episode with a thought experiment.
0:00:41 What would you do if you knew you were going to die today?
0:00:45 And what would you do if it were impossible to die today?
0:00:54 I’d bet that separate from the outcomes that those days would look pretty different because probabilities guide our decisions.
0:00:57 And that is the premise of today’s conversation.
0:01:04 We as humans have extended the aging curve, meaning we are probabilistically living longer and a whole lot longer.
0:01:09 In 1965, the most common age of death was in the first year in the UK.
0:01:12 Now that’s flipped to 87.
0:01:16 And contrary to popular belief, we are still extending lifespan.
0:01:24 So what happens when the calculus is no longer avoiding dying young but perhaps striving to live extremely old?
0:01:26 Plus, what does a life worth?
0:01:33 And how does this impact our society, not just in the health system, but the financial sector, career planning, and even direct design?
0:01:39 Joining us to discuss exactly this are Dr. Andrew Scott, author of The Long Jeopardy Imperative,
0:01:46 a book published earlier this year alongside founding general partner of A16Z by Own Health, Vijay Ponte.
0:01:49 Let’s get to it.
0:01:55 As a reminder, the content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice,
0:02:02 or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund.
0:02:08 Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast.
0:02:14 For more details, including a link to our investments, please see a16z.com/disposures.
0:02:21 You could have named the book Many Things.
0:02:27 The longevity imperative is very specific and very telling, I think, of what message you’re trying to get across.
0:02:30 Let’s start there. Why use the term imperative?
0:02:37 Partly, I’m a macroeconomist and I think we’ve got to sort of face the challenges we have in the world ahead of us.
0:02:44 And this one just drops off the radar. We talk a lot about AI and climate change now we have to adapt and adjust.
0:02:52 But when it comes to this topic, we tend to talk about an aging society and we rarely get beyond talking about adult diapers and care homes.
0:02:55 So I wanted to just sort of elevate this topic right up there.
0:02:59 And the meaning of imperative is vital importance.
0:03:08 And actually, I think that’s got to double play too because it’s not just vitally important for society because we want to avoid a pensions crisis and a health crisis.
0:03:15 It’s vitally important for all of us because for the first time ever in human history, the young can expect to become the very old.
0:03:19 And so how we age is now an incredibly important thing.
0:03:35 Let’s quickly double click on that because you actually compare the criticality of longevity to things like let’s say AI or climate change, other areas or sectors where people widely address as really critical and widely impactful.
0:03:38 Why do you position it there relative to those other areas?
0:03:42 Because I think it’s right up there just as important.
0:03:45 We fear getting old. We fear outcry. Our skills are health.
0:03:51 And so that’s why there’s that longevity imperative. What are you going to do now to age well?
0:03:58 But aging is not a sexy word. I know no one who says, “Yeah, great. We got an aging society,” which is sort of strange.
0:04:04 It’s one of the greatest achievements of the 20th century to get global life expectancy above 70 phenomenal.
0:04:10 Fewer children to mourn, fewer parents snatched away in midlife, more grandparents meeting their grandchildren.
0:04:15 And we say, “Shit, we got an aging society.” I mean, it’s just a really weird way of framing it.
0:04:21 And so people switch off with aging, whereas when it comes to AI and climate change, it’s like, “Oh, this is really important.
0:04:27 If we don’t do this, we’re going to have a bad outcome. It’s going to be crucial for humanity, crucial for my individual future.”
0:04:31 And all of those statements hold for how you deal with your long life.
0:04:35 So that’s why it’s up there. And it’s not just, I think, me being obsessed with this topic.
0:04:41 I can give you very, very hard data about the welfare gains, but also the economic consequences.
0:04:47 If we don’t adapt the pensions crisis, the unsustainable health burden, it really is right up there.
0:04:53 I don’t think anyone really, other than people who’ve looked at the economics, understand the impact this is having on us,
0:05:00 not just as individuals, but as a country and the amount of debt, the amount of expense that comes along with our current healthcare system,
0:05:05 especially which is exacerbated with age. It’s a very hidden, insidious problem.
0:05:08 And so I think imperative is spot on.
0:05:15 Absolutely. And I’d love to dive into how we do need to refashion our economic system, our tax system,
0:05:18 the way that people think about their careers, because you talk about all of that, Andrew.
0:05:25 But before we get to that, let’s maybe take stock of how lifespan and then also health span have changed,
0:05:31 since those are not the same thing. Ground us in how things have changed, maybe over the last 100 years or so.
0:05:38 Yeah. So if you look at the sort of high income countries, the last 100 years, life expectancy has increased about two or three years every decade,
0:05:45 and in lower income countries, even faster. And so global life expectancy is now over 73.
0:05:51 I think it’s worth just pausing at that point. Global life expectancy is over 73.
0:05:56 Around the world, a child born today can now expect to live into an eighth decade.
0:06:04 And actually, that’s based upon no further change. So it’s probably a greater number than that because life expectancy is increasing in many countries.
0:06:11 In high income countries like the UK, the UK government says that 50% of children born today will live to be about 91.
0:06:18 That’s extraordinary. And I start my book with a somewhat dramatic statistic, which is not about average life expectancy,
0:06:27 but what is the most common age of death? And when I was writing the book, I wanted to sort of make it personal and illustrate these demographic trends with my own family.
0:06:33 And I discovered my own family was stunningly average. They hit nearly every single demographic trend completely.
0:06:39 But the most striking one was that when I was born, the most common age of death was children under one in the UK.
0:06:46 And I was a twin and my twin died in the first few days of life. So I thought that was quite rare, but that was the most common age of death.
0:06:51 And today, the most common age of death, I think it’s 87, 88, something like that.
0:06:52 Yeah.
0:06:59 In my lifetime, the modal age of death has shifted. I mean, it’s complicated life expectancy that’s average, mean and median.
0:07:06 But wow, Sam Beckett, the Irish playwright, says we give birth to stride the grave, the light flickers instant, and then it goes out.
0:07:10 Well, the light’s flickering a lot longer now, and that’s what we’ve got to prepare for.
0:07:15 Yeah. I mean, you shared so many interesting statistics. I’ll just share one more. That surprised me.
0:07:26 Based on 2020 data, you said that a newborn girl in Japan has a 99.6% chance of living to 20 and 99% chance of reaching 40 and a 96% chance of making it to 60.
0:07:36 I mean, those are just astounding numbers, which to your point, we’re not millennia away from those statistics, like just the opposite of those a few decades ago.
0:07:40 You said in 1965, we were looking at completely different numbers.
0:07:49 And what’s sort of happening is that there’s a concept of the survival rate, the probability of a newborn reaching a certain age, and it’s sort of rectangularizing.
0:07:56 As you just said, the chances of a child, a female in Japan reaching 60 is now incredibly high now.
0:07:59 And now the chance of getting 70 is getting higher and higher and higher.
0:08:04 So that’s the real sort of shift. We’ve lowered infant mortality dramatically.
0:08:09 We’ve made big steps in improving midlife mortality, although the US has got some problems there of late.
0:08:13 And so, of course, now all the gains are coming at older ages.
0:08:24 And so life expectancy is now being driven by falls in mortality at older ages such that 80% of life expectancy gains in the richer countries is coming from mortality rates coming down after 70.
0:08:28 So it’s just getting further and further along this rectangularization.
0:08:30 Absolutely. And Vijay, I’d love to bring you in here.
0:08:37 As you look at both the bio and health side of things, what gives you confidence that this trend might continue?
0:08:42 Yeah, I think there’s a lot of excitement about studying the biology of aging for a variety of reasons.
0:08:48 I mean, first off, it’s actually kind of shocking how poorly it has been studied and how little time has gone into it.
0:08:59 But the other thing that maybe is at the heart of aging and disease is the fact that many diseases are exacerbated with age, like cancer, Alzheimer’s, type 2 diabetes.
0:09:09 And so given that there may actually be some common precursors that are interesting to go after, either from a therapeutic point of view, but obviously also from a lifestyle point of view.
0:09:14 Even simple things like we know being overweight is a massive co-morbidity for any of those things.
0:09:24 And so part of the challenge will be as a society what we want to handle through the traditional health care model, which is basically a sick care model dealing with disease.
0:09:31 Or what can we do more broadly to actually get to longevity and never sort of engage with the medical model in that way?
0:09:33 Yeah. And of course, you said it’s that through your whole life.
0:09:38 I think for me, that’s why I stress the longevity society, not an aging society.
0:09:41 An aging society says, what do we do with all these old people?
0:09:44 Whereas the real change is the young can expect to become old.
0:09:52 When you’ve only got a 10% chance of making it to 90, you really shouldn’t say think in your 430s and 40s, what can I do now to make sure I’m healthy in my 90s.
0:09:56 But when there’s a 50% chance of making it, my goodness, you’ve got to start doing it.
0:10:02 And we have this diseased focus model, which just doesn’t work as we get older.
0:10:06 Because as VJ says, chronic diseases have this common source.
0:10:08 And I think that’s really interesting.
0:10:11 What is the X that affects all these later diseases?
0:10:13 And it could be aging.
0:10:15 It could be obesity and GLB1.
0:10:17 It could be through exercise.
0:10:21 And of course, that focuses on a really key issue, which is health.
0:10:26 It just strikes me as stunning that we don’t just say, how do we maintain health?
0:10:28 We say, oh, how do we prevent disease?
0:10:31 It just shows you the orientation of the system.
0:10:38 And there’s a quote by W.A. Jordan, which I use in the book, which is sort of health is the concept that the medicine knows nothing about.
0:10:43 And I said this the other day to a cardiologist, he said, no, health is just an incomplete diagnosis.
0:10:46 Something happens and then I’m in, but it is quite striking.
0:10:48 Can we speak to that really quickly?
0:10:54 Because something that you and many others have pointed out recently is this trend toward specialization in health care.
0:10:59 And at face value, someone who’s not involved in this world, actually, that sounds pretty great, right?
0:11:07 The fact that for anything that I run into, there’s going to be some specialist who studied this much longer than the average doctor who can solve my problem.
0:11:13 And so just as we’re kind of addressing the state of affairs, can you speak to that specific trend in specialization?
0:11:21 Now, maybe it’s not as compelling as you might think if we’re trying to achieve this idea of longevity.
0:11:23 There’s always a case you made for everything.
0:11:25 I don’t want to sort of criticize specialization.
0:11:34 I think when we come to drug development, we can say that there’s some challenges there because we are seeing a lot of drugs being developed for very serious specific conditions where you can charge a lot,
0:11:37 but you only affect a small number of people.
0:11:39 Brilliant that you affect those people.
0:11:42 But what about the cross-cutting breakthroughs that we really need?
0:11:45 But no, I think it’s a consequence of this sort of disease specialization.
0:11:58 And in particular, as you start to have an older population, I think the real problem of specialization is you’d be spending one morning with a cancer specialist, one morning with a cardiovascular specialist, one with a diabetic clinic.
0:12:00 So you’re going to get massive polypharmacy.
0:12:07 So you have multiple description of drugs where you’ve got no idea how they interact because we haven’t got clinical trials to try and prove it.
0:12:14 And then I think the other challenge we’ve got is that the trouble as you get older is you get all these aging-related diseases.
0:12:18 And fantastic if you can reduce the risk of cancer.
0:12:19 That’s tremendous.
0:12:22 But you’ve still got dementia out there.
0:12:26 You’ve still got diabetes and cardiovascular, all of which will lower the quality of your life.
0:12:36 So if you could do something early that affected multiple diseases, the whole is greater than the sum of its parts, which specialization is not going to give you.
0:12:48 I think Andrew hinted in that the view that medicine starts with disease is part of the problem because it should start with health and to maintain health rather than to fix disease.
0:12:52 Like we all don’t want our houses to be on fire, so we have smoke alarms.
0:12:57 And if we wait until the house is on fire, that’s both unfortunate and also there’s only so much you can do.
0:13:02 In some ways, we ask a lot from acute care and they can do miraculous things.
0:13:09 It’s kind of amazing, but it would be a lot healthier, but also a lot cheaper to be able to address these issues early.
0:13:13 But that has its own challenges and part of it is just not the way the system is built to work.
0:13:20 I think that is the problem because what you’ve got then is you’re keeping people alive, you’re not keeping them healthy.
0:13:21 And that’s the real challenging.
0:13:25 We’re spectacularly successful in keeping people alive.
0:13:31 But as we’re going beyond 70 into 80s and 90s, we slow down the dying process, but not the aging process.
0:13:39 And that’s been a problem because you’re going to spend ever more money on sensible things, but you’re getting not better health outcomes.
0:13:45 In as an economist, there’s various tools where you can put dollar values on health gains and you can add those to GDP and say,
0:13:47 “Look, here are the welfare gains we’ve got.”
0:13:54 And suddenly over the last decade, you’ve seen a really big slowdown in the rich countries in terms of the health benefits they’re getting.
0:13:57 But health expenditure is still rising.
0:13:59 And they’re called new assets.
0:14:03 The health outcomes have been spectacularly bad compared to other countries.
0:14:10 This is a good point to really double click on the fact that I think a lot of people don’t realize the societal impact of just that spend.
0:14:19 There’s been a lot of talk we’re in an election year about inequality between middle class and upper class that a lot of middle class gains and wages have been zero.
0:14:20 They’ve been flat.
0:14:25 But actually, if healthcare could have been kept at the same cost, let’s say 30 years ago,
0:14:27 then the middle class would have seen huge gains.
0:14:31 Basically, healthcare and its rising costs is eating up these gains.
0:14:34 And so it’s not just about keeping us healthy, which would be great.
0:14:37 It’s about the health of the nation and the health of the economy.
0:14:39 So it has really broad reaching impacts.
0:14:40 It does.
0:14:47 And there’s also another link as well, which is something I’m particularly keen on at the moment, which is making a broader case of prevention,
0:14:52 which is that if you could stop people in their 50s having heart attacks, they can carry on working.
0:14:56 And that then generates huge benefits in GDP and taxes and benefits.
0:15:00 And this, of course, is the key thing because you’ve got more and more older people.
0:15:04 If you want to make them work for longer, you’ve got to keep them healthier and productive for longer.
0:15:08 You can’t just say, “Hey, I’ve raised the social security age. Good luck.”
0:15:14 So if we think about these incentives and how these healthcare systems really need to think about where they allocate their resources,
0:15:22 it really is a very multivariate equation. And I’d love for you to break down how these systems actually do decide where to invest.
0:15:29 I know there’s different metrics, whether it’s VSL or Qualey Quality Adjusted Life Year that different groups use.
0:15:35 But then in addition to that, I just wanted to call out one other metric from your book, which is about this idea of action versus prevention.
0:15:43 In 2020, total U.S. healthcare expenditure was $4.1 trillion, which is around $12.5,000 per person.
0:15:48 But of that, only around $363 was spent on prevention.
0:15:52 So I think that really highlights exactly what both of you just spoke to,
0:15:59 which is the idea that we’re not investing in prevention at least compared to the action when something goes wrong.
0:16:03 So could you just speak to how these healthcare systems choose to divvy up the resources?
0:16:07 So what’s interesting globally, we’ve got the same problem. We’ve got a pensions problem and a health problem.
0:16:12 And then of course, if you look within the details of the pension and health system in every country, it’s wildly different.
0:16:22 They’re all going wrong in different ways, but in a lot of them, basically, broadly speaking, the Ministry of Finance hand over a check to the Ministry of Health.
0:16:29 The Ministry of Health’s got that stock of money, it has to decide how to allocate it. And it will look at some health measures.
0:16:39 Again, it varies in my country, the UK, one of the big targets is waiting list for operations, which is just a crazy metric because I can’t really think how that links into health measures very well.
0:16:46 But that’s a big measure. But there’s also this medical concept of a quality, which is a quality adjusted life year.
0:16:58 And so if you have a treatment that saves a child’s life and they can live to 80, you’ve saved 80 years, but you quality adjust the years for the sort of level of health that you get.
0:17:07 And so broadly speaking, you calculate that and then you set a price for every quality in the UK, it’s around £30,000, so it’s about $40,000.
0:17:14 And say, if this treatment comes in less than that price, it’s worth it. If it’s not, don’t do it.
0:17:20 In the US, the budgets are much bigger and it’s more of a commercial decision, I would say, that happens.
0:17:24 And of course, that’s that question of how much you can pass on to the insurer.
0:17:28 But there’s all sort of value based pricing in that as well.
0:17:37 Most of the measures are all very internal to the health system. I think that’s a, for me, a really big issue because with budgets under constraint,
0:17:43 take the UK where the health system actually, they spend a lot more money, but they’re not getting better outcomes.
0:17:47 You’ve got more and more older people who need operations, need the help replacement, etc.
0:17:50 And you can’t say to them, no, I’m not going to do it.
0:17:56 So you haven’t got the money to finance prevention and you’re just stuck forever doing hit replacement operations.
0:18:01 So the case for prevention at the moment is sort of made, will it save me money within the health system?
0:18:05 If I spend this now, will it reduce my operating costs now?
0:18:10 And the problem with that is prevention tends to take a bit of time to come through.
0:18:19 The other problem you’ve got in the US health system is that with multiple health providers, one person’s expenditure could be someone else’s benefit rather than them.
0:18:26 In the UK, it’s a lot easier because we have one health system, the Ministry of Finance pays for everything so they can see the savings.
0:18:34 But this is why I’m very keen to sort of say, oh, hold on a minute, though, there’s another game to prevention, which is these spillover benefits.
0:18:41 In the UK, and I’m sure it’s similar in the US, if you’re 15, you have a heart attack, you’re six times more likely to leave the labor market.
0:18:44 And once you’re out the labor market, it’s very hard to get back in again.
0:18:51 So if you can stop that person leaving, you’ve got this knock-on effect in terms of income, GDP and revenue.
0:18:53 And that’s not currently taking into account.
0:19:01 Very little time is discussed on prevention, nutrition, and also those fields are complex and highly debated.
0:19:04 So there’s the education part, which I think there’s huge room for improvement.
0:19:08 But then, at least in the US, the big question is who’s going to pay for it?
0:19:18 And with existing payers, they unfortunately have to deal with a reality where people may switch payers every two or three years, and prevention is a long-term game.
0:19:26 There’s a big push for value-based care, and I think there’s new models for that and exciting things happening, especially in areas like Medicare Advantage.
0:19:36 But in the end, I think what we’re going to see is high deductible plans and choices like ICRA that will allow actually consumers to be the ultimate payer.
0:19:43 And in the end, we kind of are the ultimate payer in dollars, and we are obviously the ultimate payer in healthcare results.
0:19:45 And so we care about value.
0:19:50 I care about my children’s health and my wife’s health and so on in a way that no insurance company would.
0:19:57 And I think that shift for who’s going to pay and me choosing how my dollars are spent could also facilitate more prevention.
0:20:04 And it’s a really interesting point too, because of course, once you move away from the disease model, the health system and now everything, it’s the air that you breathe.
0:20:07 It’s the British state that you live in. It’s the food that you buy.
0:20:10 And I think you’re going to see it go retail in a very big way.
0:20:13 Of course, not everyone is going to spend and invest in health.
0:20:16 But I do think that’s the logical consequence of prevention.
0:20:23 If the health providers won’t provide it sufficiently, other people will, whether that be wearable devices or whatever.
0:20:26 So I think the health economy just starts to get even bigger.
0:20:32 And I think something that a lot of people don’t realize is that if you look at the determinants of health and especially mortality,
0:20:35 genetics is a large part, it’s like 30 or 40 percent.
0:20:38 All of medicine is basically 10 percent.
0:20:42 And then a large fraction like 40 percent is social determinants.
0:20:44 So if your spouse smokes, guess what?
0:20:49 You’ll either be smoking or you’ll have secondhand smoke where your zip code predicts your health to a large degree.
0:20:54 Those social determinants are a huge part and addressing that should be part of health care.
0:21:01 But right now, the health care really is that 10 percent of providing disease focused care and you’re missing that 40 percent.
0:21:02 I completely agree.
0:21:04 I mean, the socioeconomic determinants are huge.
0:21:07 So behavior and environment explain the majority.
0:21:11 But one thing that’s interesting is people kind of resistant to the idea that aging is malleable.
0:21:15 But you just look at inequality and you say, well, we found a way to speed up aging.
0:21:21 These socioeconomic determinants, so it’s proof positive that we can really affect how we age.
0:21:24 And it somehow just doesn’t get into the consciousness.
0:21:29 Yeah. And I think examples that you gave in the book are that on a very small scale,
0:21:35 you are seeing some insurance companies incentivize people who exercise or have healthy habits or get discounts on healthy food.
0:21:38 I’m surprised we’re not seeing that more candidly.
0:21:46 But maybe this is a good transition for us to talk about social implications and the major parts of society that do need to be restructured.
0:21:49 As some of this is somewhat inevitable, right?
0:21:54 There’s things like the health system, the financial sector, taxes and career planning, drug design.
0:22:01 Yeah. And let me just reiterate the very simple idea of the book, which is for the first time ever, you can now expect to become old.
0:22:04 So you’ve got to make sure that you’re healthy and engaged for longer.
0:22:06 You’ve got to invest more in your human capital.
0:22:10 And of course, if you don’t invest in your future, then actually your future isn’t very good.
0:22:13 You run out of money, you run out of skills, you run out of purpose, you run out of health.
0:22:21 But we had never set up our institutions to say, how does Andrew live a life to his 90 that remains healthy and engaged?
0:22:26 So nothing is sort of untouched, including culture and psychology.
0:22:27 But where are the main things?
0:22:29 Well, obviously the health system is huge.
0:22:35 I mean, a shift to a focus on delivering health rather than treating disease.
0:22:37 Work clearly has to change.
0:22:45 And in an early book, The Hundred Year Life, I sort of talked about how we created a three-stage life in the 20th century of education, work, retirement.
0:22:50 But if you just stretch that out to 90, it looks pretty unappealing, pretty miserable.
0:22:55 And I think for me, the very simple notion here is longevity is about having more time.
0:22:58 So the question is, what do you want to do with that time?
0:23:01 And how do you in particular distribute leisure across your life?
0:23:07 And in the 20th century, we fixed retirement age, lived for longer and just took more and more leisure at the end of life.
0:23:09 Governments around the world are saying, no, that doesn’t stack up.
0:23:11 You’re going to have to work for longer.
0:23:16 So I think we may take less leisure at the end of life, but we’ll spread more leisure across life.
0:23:18 We might start work later.
0:23:20 We might take mid-career breaks.
0:23:24 Your last job might be part-time or you might go part-time sometime in the middle.
0:23:30 But your career is going to have, I think, a lot more transition, some chosen by you, some forced upon you.
0:23:34 And that, I think, has big implications for all sorts of different people.
0:23:39 So if you’re a manual worker age 50, you can’t carry on working in a physical job.
0:23:42 How do you transition into something different?
0:23:48 If you’re an investment banker working all those hours, you can’t carry on doing this beyond the age of 45, 50.
0:23:51 Do you take a break, retrain and do something?
0:23:55 And wrapped into this is education, identity, so many things.
0:24:03 But finance comes into it as well because the simple pension industry says, accumulate money while you’re working, run it down when you retire.
0:24:09 But if you’ve got this sort of multi-stage life, accumulation to accumulation becomes much more complicated.
0:24:12 So I think finance has to probably do three things.
0:24:17 One is do long-run saving products that have a much more flexible pattern.
0:24:20 Totally tie financing into health.
0:24:29 And then the other big challenge is that, you know, in the 19th, 20th century, we developed a life insurance industry that paid out if you died young.
0:24:34 But sort of the risk now is not that you die at the average, but most people die above the average.
0:24:41 And there’s now a small chance of going above 100, one in five, one in six Brits, and then to live to 100 according to the government.
0:24:43 How do you finance that?
0:24:52 So that living insurance, how do I provide you a steady stream of income in case you sort of live a really long life is a huge financial one?
0:25:01 Yeah, I think there’s a crisis also that those who have anticipated see it, but I don’t think everyday people think about it this way, which is the shift at least in the United States from pensions,
0:25:06 which have defined payouts to 401ks, which have defined contributions.
0:25:13 That in the context of living to 90 and maybe in time to 100, is your 401k set for you to live to 100?
0:25:15 Is it built that way?
0:25:17 And so what does that mean for when you stop working?
0:25:23 And can you have a health span long enough so you can build a 401k so you can pay for your lifespan?
0:25:26 All these things are things that are going to be fundamental to us.
0:25:29 And by the time we’re there, it’s too late to make changes.
0:25:31 Yeah, and it’s a really complicated problem isn’t it?
0:25:33 You think how long will I be able to work for?
0:25:35 How long will I live for?
0:25:38 How much money do I need my retire and what’s the rate of return going to be?
0:25:41 Those answers are going to differ wildly for different people.
0:25:43 It’s impossible to be precise about them.
0:25:47 So whatever plan you’ve got, you’ve got to have some flexibility in it.
0:25:53 And the key flexibility is investing in your human capital, investing in your health and your skills.
0:25:57 So if suddenly you find you haven’t got enough money, you can carry on working for longer.
0:26:01 But if you haven’t got the skills and haven’t got the health, then you’ve got a really big financial problem.
0:26:03 So I think that’s the other thing about longevity.
0:26:07 The portfolio suddenly becomes not just my 401k.
0:26:09 It is my health and my skills.
0:26:13 And integrating those together is really important.
0:26:18 And I think something you’re calling out is just how this dynamic has completely flipped on its head, right?
0:26:21 Before it was, how do I avoid dying early?
0:26:24 And now it’s, what do I do if I live too long?
0:26:26 That is a really interesting phenomena.
0:26:30 But to your point, the statistics back it up in terms of your probability.
0:26:33 And we do make decisions as humans based on probability.
0:26:38 Maybe on that end of the people who maybe might say, you know what?
0:26:43 I don’t know if I want to live forever or living to 100 doesn’t actually sound so great.
0:26:50 You mentioned pretty repetitively throughout your book this bias that people seem to have around aging and the aging population,
0:26:59 both on a personal level, but then on the societal level of requiring the support of many others who have not sustained their health span.
0:27:04 So can you just speak to this further and talk about how maybe we need to update our priors based on these shifts?
0:27:07 There are definitely some cultural differences here.
0:27:13 China and the U.S. is a one end of the spectrum, but in general, we underestimate the capacity of older people.
0:27:14 That’s ageism.
0:27:18 And the trouble with that is it means we underestimate the capacity of our own later life.
0:27:20 And I enjoy telling a story.
0:27:26 I was teaching some Chinese MBA students this summer and I was showing them the Chinese demographic data.
0:27:29 These students are late 20s, early 30s.
0:27:32 And China’s demographic changes is truly striking.
0:27:38 It’s going from 1.4 billion people to 1 billion, 45% in the movie more than age 65.
0:27:44 So I was sort of showing the data and showing how in 35 years time there’s all these people aged over 65.
0:27:47 And I said to these Chinese MBA students, how do you feel about it?
0:27:48 And they said, it’s a problem.
0:27:49 I said, why?
0:27:52 They said, but there’s all these old people, they’re going to get ill, they need a pension.
0:27:54 And I said, well, who are these old people?
0:27:56 And they said, what do you mean they’re old people?
0:27:57 I said, no, they’re you.
0:28:01 And it was remarkable how the penny dropped like, oh my goodness.
0:28:05 When everyone hears about an aging society with old people, that’s not about me.
0:28:06 I’m going to be fine.
0:28:07 I’m going to be fit and healthy.
0:28:10 And it was quite remarkable how we’re negative about old people.
0:28:14 And we don’t recognize that the big changes the young can expect to become the old.
0:28:18 I think there’s all sorts of reasons for that about why we have that negativity.
0:28:23 But it’s a fundamental problem because you underestimate the capacity of your own later years.
0:28:24 You will under invest in them.
0:28:28 And so you lead to the very sort of outcomes that you fear.
0:28:32 And then there’s just all the sort of problems of ageism that comes with any form ofism,
0:28:37 which is that if you assume that on the basis of a characteristic age, this is what you’re like,
0:28:41 then you’ve got a problem because actually the real thing about aging is diversity.
0:28:43 People age really, really diversely.
0:28:46 And you can be a hundred and literally running the hundred meters.
0:28:50 You can look at it on YouTube or you can be 50 and in a wheelchair.
0:28:54 And I think society has become very focused on chronological age.
0:29:01 I think it’s really interesting in England in 1601, we introduced the Elizabethan poor laws.
0:29:04 And it was look after the poor and the old.
0:29:06 But no one knew how old people were.
0:29:08 There weren’t birth certificates.
0:29:10 There wasn’t much numeracy.
0:29:14 So until very recently, people didn’t know when they were born or how old they were.
0:29:18 The song “Happy Birthday” comes in the 1930s, just kind of.
0:29:25 So what the poor laws said was, okay, old is someone who’s lived a long time and can’t look after themselves.
0:29:28 So the whole test was about, could you be functional?
0:29:30 Then bureaucracy comes along.
0:29:32 So this is a nightmare, this test.
0:29:34 Let’s just say everyone over 65 is old.
0:29:36 And that’s what we do today.
0:29:38 We say everyone over 65 is old.
0:29:43 But of course, the problem that chronological measure of age is measures how long you’ve lived.
0:29:46 And for me, there’s two much more important measures.
0:29:49 One is how many more years you can expect to live.
0:29:50 I’m 59.
0:29:56 And at 59, I have to behave differently from my father and my grandfather at 59 because I can expect more years.
0:30:02 And then of course, the other thing is my biological age, sort of how is my body aging and how am I dealing with it.
0:30:08 And of course, if you have that approach, you start to think, well, actually being 59 doesn’t really tell me much about Andrew.
0:30:11 It’s not that important a piece of information.
0:30:16 But we zoom in on chronological age, we assume everyone over 65 is old.
0:30:19 That is an enormous problem for firms, for instance.
0:30:24 Because whenever I sort of see a presentation of the aging society store and a consultant comes on,
0:30:28 I know immediately they’re going to say, hey, we should invest in care homes and cruise ships.
0:30:33 Because everyone over 65 is going to want a cruise ship or go to a care home.
0:30:36 And you lost the market completely, if that’s the way you think.
0:30:44 Yeah, I was just right before this actually watching a video of Robert Marchand, the French 105 year old cyclist.
0:30:50 He unfortunately passed away at 109, but at 105 he was still cycling, still beating centenarian records.
0:30:56 And so to your point, age very much isn’t a metric, but there are many others to pay attention to.
0:31:04 And so I feel like a lot of listeners at this point are convinced that maybe longevity is more complex and important than I realized prior.
0:31:08 And so on a personal level, they’re probably asking, OK, so what should I do?
0:31:13 Because I can impact this. Maybe starting out there, how would you both think about the 80/20?
0:31:17 The big problem is both when we talk about diet and exercise.
0:31:23 I think the first misconception is that there is a right diet for everybody and a right exercise plan.
0:31:31 And this is so individualized that we’ve seen now lots of companies crop up to try to measure things that will allow you to know what the right diet is for you.
0:31:35 And so it starts with diet and exercise, but I think that’s a lot easier said than done.
0:31:38 It’s really trying to figure out what is important for you.
0:31:45 But from there, I think there are intriguing things on the horizon that are maybe probably more a little bleeding edge.
0:31:48 People take metformin, people take other supplements.
0:31:52 I put that in the 20 or the 5%. I wouldn’t start with that at all.
0:32:01 I think if you’ve optimized your diet and you’ve optimized your exercise and you’re still getting PRs in your 50s of either weights or cardio or whatever,
0:32:07 and you want to go that last bit, that’s one thing, but I think there’s a lot of low-lying fruit, especially for people who are fairly sedentary.
0:32:15 I’m not a medical doctor. I don’t take anything as not prescribed and proven simply because I can’t monitor my own health to know if it’s working.
0:32:20 I was saying this the other day, someone takes lots of supplements and they were shocked and said, “We’ll miss you, Andrew.”
0:32:24 I don’t take any of that and I agree there’s exciting stuff coming along.
0:32:28 It’s always a disappointing thing because people always want the easy answer and there are some easy answers.
0:32:32 It’s exercise, diet, sleep, enjoyment, good company.
0:32:40 At the moment, I would broadly say, although there’s some latest evidence about high-intensity exercise and this way of sleeping, et cetera, and fasting,
0:32:45 broadly speaking, we know what works, what’s changed is our incentive to do it.
0:32:51 I think that’s the message that you’ve got to draw home, not that there’s some secret technique that’s going to change everything.
0:32:57 We have done brilliantly inventing and innovating intervention techniques.
0:33:02 We’ve now got to put the same thing on to prevention and data is going to be really important.
0:33:07 Then I think we’ll start seeing some more personalized medicine coming along.
0:33:14 None of that is going to happen quickly, so we’re left with the stuff that really does make a big difference.
0:33:24 If you look at the evidence on particularly healthy life expectancy, you’re talking about another year or two for each of the things you do in terms of weight exercise, that stacks up quite a lot.
0:33:34 I’ve seen myself and others, it’s not overnight, but you can see gains in the first year and then they compound and then it’s about keeping it going.
0:33:39 I think the maintaining is the hardest part to avoid injury and to build that discipline.
0:33:42 We could also talk a little bit about the sci-fi stuff.
0:33:53 Beyond the 80/20, this is like the 99.9 to 0.1, but the sci-fi stuff is that there are just amazing new discoveries in biology of cells and even organisms.
0:34:00 The so-called Yamanaka factors and many people are searching for what really could be a fountain of youth.
0:34:04 That is way off and I don’t think you’re going to be taking a pill like that anytime soon.
0:34:12 But intermediate things, there’s new pathways that are corresponding to pathways of exercise and there’s drugs in phase two for that.
0:34:18 Even the GOP ones and there’s pros and cons of them may help you get started if you’re obese.
0:34:24 It’s about the coming up with a plan where this is not going to be a fad diet, this is going to be the rest of your life.
0:34:32 The GOP ones are really interesting and I think we’ve still got some way to go to work out exactly what they’re doing.
0:34:39 But what is so interesting is first of all, it looks like it gets people a sense of agency and control so they can do other things.
0:34:41 So it has benefit mental health.
0:34:49 It’s affecting multiple diseases. It’s a bit like one of those biomarkers and some of the effects seem to happen outside of an effect on obesity,
0:34:55 which of course is exactly the sort of thing that we need to have because the body is very complicated.
0:35:00 Aging is not just one process and so there’s unlike to be just one pill we take that does everything.
0:35:06 But I give the analogy with interest rates and the economy, the government changes interest rates to try and control the economy.
0:35:09 It’s not perfect, but it has a pretty big impact.
0:35:14 And so what’s the equivalent and I think that’s a really interesting example of the GOP one.
0:35:16 And then as you say, I think it’ll be lots of other stuff.
0:35:21 I’m particularly taken by stem cells, which you can then just regrow a liver and that’s pretty sci-fi about that.
0:35:27 I think it’s sort of pretty feasible to happen or drugs that treat arthritis. Wouldn’t that be wonderful?
0:35:35 So I think that’s the sort of beginnings of aging better rather than here’s something that’s going to keep us alive to 150.
0:35:44 Yeah, if I’m taking anything away from both of you, it’s just that maybe we are at this interesting juncture where it really pays to think about this.
0:35:54 And I know that sounds really simplistic, but if we take your example, Andrew, of your kind of family tree, that longevity that happened decade over decade was a little bit passive.
0:35:59 It wasn’t passive in terms of there was millions, if not billions of people working toward those longevity gains.
0:36:06 But your father or your grandfather weren’t necessarily thinking, “Let me go test my VO2 max. Let me eat certain things.”
0:36:12 But now we do have more information than ever, especially in, as you said, Vijay, that 80%, right?
0:36:15 We’re not even talking about the sci-fi stuff here.
0:36:19 So there are very clear things that people have agency over.
0:36:28 Are there any other second, third order effects of this “long-jevity revolution” that you don’t think we’ve necessarily touched on yet that really people should be thinking about?
0:36:31 For example, one that you mentioned in your book, Andrew, was menopause.
0:36:39 There’s something that a lot of people don’t think about, but the health impact to 50% of humans is pretty substantial.
0:36:47 Menopause is very interesting because of course there’s a huge impact on women’s later life and women tend to live for longer, but get into ill health before men.
0:36:51 And so successful menopause is really important for later future health.
0:36:56 And it’s an accelerated form of aging and it’s quite rare in the amongst animals.
0:36:58 So humans are quite unusual in having a menopause.
0:37:05 And of course, men’s reproductive ability declines in a general form of age, but women’s doesn’t.
0:37:07 So I think that’s a really interesting area.
0:37:16 And our role models about how we live and what is old and what isn’t old have been formed from centuries and millennia and they radically need to change.
0:37:26 And then the other thing I think it’s important to stress is that we talk about it being more old people, but in many countries, the U.S. as an example used to have a pyramid that was lots of people at the bottom.
0:37:28 Now it’s more like a straight tower.
0:37:30 And so you’ve got age equality.
0:37:36 And because of that intergenerational connectivity becomes incredibly important.
0:37:44 And we design our institutions around a hierarchy, but we’ve got to get much better at that intergenerational mixing.
0:37:48 How do you stop a hierarchy blocking progress for the young?
0:37:52 How do you generally learn from one another and exploit that?
0:37:54 That’s going to be incredibly important.
0:37:57 There’s a virtuous cycle and a vicious cycle.
0:38:02 The vicious cycle is what we’ve been talking about where people are not paying attention to their health.
0:38:08 They can’t work, but they live long enough that they’re very expensive and not contributing to GDP.
0:38:12 The virtuous cycle is where people can maintain health span.
0:38:16 They can continue to work, continue to contribute to their 401(k).
0:38:28 And instead of this being a crisis, it could be the greatest accomplishment of humanity to allow people in their 60s and 70s and 80s to have massive contributions like we’d only expect from people who are younger.
0:38:30 That was an amazing future.
0:38:32 And theoretically, we have all the elements.
0:38:34 We just have to choose to do it.
0:38:36 And it’s sort of mobilizing people to this problem.
0:38:38 I think is the imperative.
0:38:46 If you think about it, the FDA was started before we even had access to penicillin or modern-day antibiotics.
0:38:55 Which I think kind of just speaks to the idea that these institutions, while they do many good things, are really, at least to some extent, a relic of the past.
0:38:58 But that’s exciting because there are clear wins.
0:39:09 Yeah, I often quote Malthus because, of course, Malthus, back in 1799 or whatever, comes up with this very miserable thesis that says populations grow exponentially, our resources grow linearly.
0:39:14 We’ve got too many people and we’re always going to have problems of illness, disease, famine, etc.
0:39:18 And that negativity is sort of shared with the aging society story.
0:39:20 The aging society story doesn’t say we’ve got too many people.
0:39:22 It says we’re just living too long.
0:39:24 We’re out living our religious support this life.
0:39:27 So we’re going to get ill and we’re going to have a pensions crisis.
0:39:31 So it’s interesting to go back and Malthus’ writing on the world population was not even one billion.
0:39:33 Now it’s over eight.
0:39:40 And what Malthus got wrong is he didn’t see innovation, invention, ingenuity, and new institutions that would come with industrial revolution.
0:39:42 And they increased productivity.
0:39:46 We’d invest in health and education, which further increased the quality of life.
0:39:49 And I think that’s a metaphor for this aging society story.
0:39:50 Where’s the invention?
0:39:51 Where’s the ingenuity?
0:39:52 Where’s the innovation?
0:39:54 Where’s the new institutions?
0:39:57 Because we can make this long life healthier and more productive.
0:39:59 We just go start doing it.
0:40:01 All right.
0:40:02 That is all for today.
0:40:05 If you did make it this far, first of all, thank you.
0:40:13 We put a lot of thought into each of these episodes, whether it’s guests, the calendar touchers, the cycles with our amazing editor, Tommy, until the music is just right.
0:40:19 So if you’d like what we put together, consider dropping us a line at ratethespodcast.com/a16z.
0:40:22 And let us know what your favorite episode is.
0:40:23 It’ll make my day.
0:40:25 And I’m sure Tommy’s too.
0:40:27 We’ll catch you on the flip side.
0:40:29 (upbeat music)
0:40:39 [BLANK_AUDIO]

0:00:05 Manufacturing is just that by a thousand paper cuts and it’s also just really, really overlooked.
0:00:12 When America was crushing it in the Apollo era, we got to the moon quick. We can’t go back to the
0:00:17 moon decades later at the same speed that we did when a computer was like the size of the room.
0:00:22 Everyone operates like we are still the industrial superpower and we’re making really poor decisions
0:00:27 because of it. There’s another phrase, “When you’re fed, you’re not hungry.” And I think this is why
0:00:32 we don’t build things fast enough in the U.S. If a carrier fleet went across Taiwan and we had to
0:00:36 ship a bunch of missiles over there and fire them, our entire inventory of every munition in the
0:00:41 country would last three days and then it would take us three years to replace it. If we don’t do it
0:00:47 now, it was going to do it 30 years later. America is a country of immense wealth, but we’re also
0:00:53 a country that’s gotten used to wire harnesses on the back of airline seats that cost $10,000 a piece.
0:01:00 Or as of 2022, building one commercial ship compared to nearly 800 comparable ships in China.
0:01:06 Or going Starliner that cost over a billion dollars and the heavily reported on failure.
0:01:10 You’ll hear more about all three of those things in today’s episode, but of course it’s not all
0:01:16 doom and gloom. There are tons of companies working toward rebuilding our industrial base,
0:01:20 including three founders you’ll hear from today, from this recording at LA Tech Week.
0:01:28 We have Jordan Black from Senra Systems, Chris Power from Hadrian, and Ryan Harjes from Castellian.
0:01:34 That was all of our Sue. Partner on our American Dynamism team who moderated this conversation with.
0:01:39 I’m Jordan Black, CEO of Senra Systems. We design and manufacture wire harnesses and probably everyone
0:01:43 in this room is asking themselves what the hell is a wire harness. So think of it as like an iPhone
0:01:49 charger but really, really, really complex, very manual and labor intensive thing to manufacture
0:01:52 25% of the cost of your car and almost every company hates building it.
0:01:58 You might also recognize Chris, who was on the podcast in 2022. We’ll link that episode in the show notes.
0:02:03 We build autonomous factories for aerospace and defense. The product that you get is either a
0:02:07 factory or a bunch of high precision components, but really we’re a manufacturer masquerading as a
0:02:13 software company, like fully vertically integrated. And finally, Brian, who’s building something
0:02:18 entirely different. At Castellian, we are actually developing hypersonic weapon systems.
0:02:24 National power stems from central cores, whether it’s economic, cultural, industrial,
0:02:28 as well as military strength. None of us at Castellian actually hope to use
0:02:33 what we’re building, but by the very fact that the U.S. has it actually acts as a deterrent.
0:02:38 So what will it take to revive the industrial base and build the next generation of manufacturing?
0:02:42 What role has regulation played here and how can vertically integrated companies
0:02:51 turn this ship around? Listen in to find out. As a reminder, the content here is for informational
0:02:56 purposes only. Should not be taken as legal, business, tax or investment advice, or be used
0:03:00 to evaluate any investment or security and is not directed at any investors or potential investors
0:03:06 in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the
0:03:11 companies discussed in this podcast. For more details, including a link to our investments,
0:03:14 please see a16z.com/disposures.
0:03:22 Why don’t we start with the present? And I know all of you have a lot of thoughts
0:03:27 about this. So expand on how you see the state of affairs right now in your field.
0:03:30 I think the one word answer is a complex. I think manufacturing is just
0:03:35 death by a thousand paper cuts. And it’s also just really, really overlooked. I’ve been in
0:03:38 manufacturing for a while and I was the kid that was made fun of in college to work at the plants
0:03:42 at Ford and it’s like, why aren’t you behind the computer? And now it’s really amazing that we’re
0:03:48 making it sexy again. But overall, it’s the most complex thing and it’s from not only processes,
0:03:53 raw materials, the design of it. But I think it’s the most exciting but also challenging problem ever
0:03:57 because it’s bits and atoms and you can only make something so cheap or so fast and
0:04:01 you need the world’s greatest minds working on it. So you look at companies like SpaceX and it takes
0:04:06 a lot of money and a lot of time to develop something that amazing. And so I think we
0:04:11 are in a really, any exciting pivotal point. But I think the one word answer is complex.
0:04:15 I fully agree that we are on the up as far as the cultural change that needs to happen to make
0:04:19 manufacturing more robust and more resilient. But we’re not out of the woods yet by any stretch
0:04:23 of the imagination. So I love to understand, Chris, your thoughts on how we got to where we are
0:04:29 and what you see as the road that took us here. Financial power is a bad proxy for military power
0:04:35 and military power is ultimately a bad proxy for industrial power. So basically the story is
0:04:41 we kicked ass in World War Two in the 1920s because we were really good at domestic manufacturing.
0:04:44 We pivoted all to the industrial base. We kicked the shit out of everyone else.
0:04:52 Even though we had worse technology, we just built more of everything. We just outbuilt everybody.
0:04:56 And what ends up happening over multiple generations is it’s like real estate traders
0:05:01 forgetting that market crashes exists. And the reason why real estate market crashes happen
0:05:05 is because there’s a bunch of 27-year-olds that don’t know the anxiety of losing their shirt.
0:05:09 And then everyone keeps clicking the buy button and it all falls apart.
0:05:13 And that happens over multiple generations, right? So what happened in the ’70s, ’80s, and ’90s is
0:05:18 we were so wealthy that generation who ended up doing a lot of financial engineering. You can
0:05:23 talk about this with baby boomers with houses or you can talk about this with cocaine in New York
0:05:29 was we stretched the rubber band really far to spreadsheets and super far away from reality
0:05:34 and started using the big stick of financial warfare and weaponry and that became like a lot
0:05:39 of America. The problem with that is it’s like being Mike Tyson in your ’50s, but you’ve got
0:05:46 some brain disease that you believe you’re still 25. So what happens is everyone operates like we
0:05:50 are still the industrial superpower and we’re making really poor decisions because of it.
0:05:56 But this happens to every great company, nation. It’s like IBM used to just make the best products
0:06:00 and happens to countries and people as well. And it’s largely because you’re so successful
0:06:04 you get abstracted away from reality and then you forget. No, the roads aren’t getting paved.
0:06:10 Like how did that happen? So the current state is bad for many reasons. One is people in the
0:06:17 government or people in power have not updated their mental models that we are super close to
0:06:23 retiring age and still pick and bar fights with 22-year-olds. It’s a really, really bad idea.
0:06:27 And then on the practical side, it’s like trying to wave a magic wand and saying we have no
0:06:32 software engineers in this country at all. Someone’s going to start open AI from a talent
0:06:37 perspective, right? And you don’t get really good AI researchers without having millions and
0:06:40 millions and millions of software engineers to train each other and build on top of these
0:06:45 abstraction layers and the whole industry gets better over a long period of time. And that’s
0:06:48 where manufacturing is at, which is why nothing’s coming out of those Intel plants for semiconductors
0:06:52 for like a decade. Because you can’t just wave a magic wand and push yourself up to the top of
0:06:58 the abstraction stack. And it really all is a cultural problem. I mean, you can ask the question
0:07:02 why are we so behind on shipbuilding? One big answer is no one cares and there’s bad government
0:07:07 incentives. The other answer is like the leadership doesn’t really know anything about shipbuilding.
0:07:13 The workforce kind of forgot to care because no one has been unimportant for such a long time.
0:07:17 The rate of what we can produce things in this country for defense or space or anything outside
0:07:23 of very rare companies like SpaceX is several orders of magnitude worse than the public believe
0:07:28 or the government believe. And our adversaries know that, which is why they’re just pushing and
0:07:35 keep boiling the frog. Because here’s an example. In most war game scenarios, if a carrier fleet
0:07:39 went across Taiwan and we had to ship a bunch of missiles over there and fire them, our entire
0:07:44 inventory of every munition in the country would last three days. And then it would take us three
0:07:49 years to replace it. That’s how bad it is. There are many answers on how to solve this, but that’s
0:07:54 the main story here. So there’s this financialization piece that led to a lot of abstraction with real
0:07:59 world power. And then there’s the culture piece, which is mutually reinforcing. We see a lot of
0:08:03 headlines about things like the environmental hurdles that SpaceX gets put in front of them,
0:08:08 or this tweet that we saw about how wire harnesses in airplane seatbacks were constantly breaking and
0:08:13 they cost $10,000 a piece to replace because of NFAA approval process. I’m curious, how does
0:08:17 regulation play into the state of affairs in making these things? And how does that affect
0:08:22 costs, timelines, things like that? Yeah, I think I kind of look at history where we had to build
0:08:29 one or two or a lower quantity of planes or missiles, and we had all these crazy requirements.
0:08:33 And just no one is taking a stance at any of these big primes saying we really should
0:08:39 stop doing that. It’s like trying to do an x-ray test on every single burr that comes out of it
0:08:43 and out. That’s how I would put the analogy together. And those wire harnesses, there’s
0:08:47 probably some guy who comes to that shop and inspects it for three hours and then makes
0:08:51 sure it goes into plane versus a shop like ours. We have a lot of automation processes to do that.
0:08:57 And so you can provide a better solution to these companies and no one is actually saying,
0:09:01 okay, we should probably lower the cost, lead time, the quality. Overall, I think these companies
0:09:05 aren’t getting the pressure. And so there’s this new age of companies like Andrew are coming in the
0:09:09 space where now I think the aerospace and defense primes are starting to get a little pressure.
0:09:15 Hopefully in five, 10 years there will be cheaper wire harnesses, but that is a ludicrous cost.
0:09:19 And I think just no one’s actually waiting the finger. Yeah, Brian, I want to turn to you for
0:09:24 a second because Chris and Jordan are building companies that are expanding our manufacturing
0:09:29 capabilities of things that exist today and that we are making just in a horribly inefficient and
0:09:33 cost-effective way. What you’re building is actually in many ways like a net new capability
0:09:40 for the country. Do you see any differences in making something that is net new in that way
0:09:45 versus making things that exist far better? So I’m going to actually build off of the previous
0:09:51 question and get to the new one all over. So when you look at not only just regulation,
0:09:56 but how do you do aerospace engineering? Why is it that traditional companies and
0:10:02 especially Boeing right now is really just eating it hard? Why is it going so poorly
0:10:08 for them? And why is the regulatory environment so harsh? My opinion is what’s happened over time
0:10:13 is with any process or experience, eventually something goes wrong. Somebody makes a bad decision
0:10:19 and they replace that bad decision with process to basically accommodate the lowest common
0:10:25 denominator. So if all your decision is making process so that you can’t do anything innovative,
0:10:30 you can’t move fast because there’s a presumption that you might screw it up which is always true
0:10:37 when you do anything. You end up going incredibly slow and now you are the lowest common denominator
0:10:42 and that’s what we have done to ourselves as a country both on the regulatory front. It’s also
0:10:47 what traditional aerospace has done to themselves in terms of how they do their internal engineering
0:10:53 culture. And so why is traditional aerospace so awful at developing new capabilities without
0:10:58 sight of certain exceptions like SpaceX and some other newer companies? Well, it’s because they
0:11:03 have 50 years of process they’ve layered on top of themselves that has literally killed their ability
0:11:09 to innovate and to move affordably and to build its scale. And this just happens with age unless
0:11:14 you actively work to keep it out. And so the most important thing whether you’re building a wiring
0:11:19 harness or whether you’re doing something new is how do you not layer on process that basically
0:11:25 turns everybody from an individual who has skills and the ability to make a mistake to just paper
0:11:30 pushing. You have to look at that at every level and in regulatory it’s the same way. It’s not
0:11:36 unreasonable that the FAA wants to keep people safe from something falling on their head but it
0:11:42 is unreasonable to take 180 days. That’s holding the progress of the country. We have to do better
0:11:46 than that but there’s a process they put in place because some point somewhere somebody made a bad
0:11:53 decision. It’s like if you filled your company with lawyers nothing’s going to get done because
0:11:58 everyone wants to keep their job and now every report coming up to the CEO is like look how many
0:12:02 legal compliance processes they put in place. So one of the reasons why people are absolutely
0:12:07 correct about government overspending is because if you have regulatory agencies with thousands of
0:12:13 people they get promoted by blocking stuff. 100% true. Yeah. It’s rare you get promoted because
0:12:18 hey I did a good thing and I saved money like especially with good this is like really bizarre
0:12:22 with the government. If you don’t spend all the money that you were given that year they literally
0:12:29 give you less as punishment next year. See what incentive structure is that? We spent so much time
0:12:35 thinking like how do we prevent becoming what we’re complaining about and it’s not easy because
0:12:40 things creep in constantly. You have to look at everything with an unbiased opinion of like why
0:12:44 are we making this decision? Why are we putting in place this process and is it more damaging than
0:12:50 what it’s supposedly fixing? I think what people miss about SpaceX and you can override me here
0:12:55 but I think people miss that a big chunk of the value of vertical integration is just not dealing
0:13:01 with a bunch of regulatory things with a bajillion different sub tier suppliers or processes.
0:13:06 Oh it’s 100% true like if you build a product and you’ve put in place a thousand legal contracts
0:13:12 with all your suppliers that have been negotiated. God untold number of… Someone’s building the
0:13:15 engine and in fact someone’s building the way. And then you’re determined later from an engineering
0:13:21 perspective that hey man if I change this this would just make everybody’s life easier including
0:13:26 your suppliers but then you have to go and renegotiate every contract you’re just like oh hell
0:13:30 with that forget it we’re gonna make unoptimal now this is the boat we’re on let’s do it and that’s bad.
0:13:34 You hit on something there around vertical integration and this is something I really
0:13:38 want to spend a bit of time on because all three of you guys are building companies
0:13:42 that can be characterized as ones where the factory is the product right to take that
0:13:47 famous Elon saying what do you see as the reasons for doing this aside from that organizational
0:13:52 culture piece from a technology perspective from a sales perspective how do you think about that?
0:13:58 We think about vertical integration is the control schedule because if you look back at
0:14:03 when America was crushing it in the Apollo era we got to the moon quick we can’t go back to the
0:14:08 moon decades later at the same speed that we did when a computer was like the size of the room
0:14:15 when they’re hand soldering discrete components onto a board how is that possible? But a large part
0:14:23 of it is because to do engineering of anything difficult or manufacturing of anything difficult
0:14:28 is a completely unreasonable request to have your engineers just sit down and design the perfect
0:14:34 system through analysis the first time through the first design is complete crap and you’re
0:14:39 going to do it again and so the only thing that’s important is optimizing for pace of learning
0:14:46 vertical integration can allow you to go so much faster because everybody’s incentives are now aligned
0:14:51 to the speed that you want especially in development and a lot of times what happens in traditional
0:14:55 aerospace is your supply chain when you’re not vertically integrated I mean hell you’ll get
0:15:01 like a new rev it’ll be a year or two years well how can you learn and how can you say oh I made a
0:15:07 mistake and I’d like to undo it without a two-year penalty you can’t it’s not like one traditional
0:15:14 aerospace program is behind schedule and over budget they all are it’s not because the people
0:15:20 are bad or stupid it is because they work in a process that just fundamentally broken and vertical
0:15:26 integration in my mind a control schedule is just what you have to do because in the polar
0:15:32 area they would launch multiple rocket flights a day to test I mean SpaceX is getting back to that
0:15:39 but I mean we had that decades ago so when you look at from an adversarial perspective China
0:15:46 North Korea Iran they can’t afford to do aerospace the way the US does aerospace what are the best
0:15:49 things that could ever happen to them is they get wealthy enough they’re like god we’ve got to do it
0:15:53 like the Americans and they just destroy their ability to go fast but right now they approach
0:15:58 aerospace the same way the US did in this in like the 60s and they’re kicking our butt they’re going
0:16:05 much faster and that’s what we have to bring back to the US my opinion one of the big thesis is for
0:16:10 our company is no one bothered to put software engineers in manufacturing for like 40 years big
0:16:16 macro statement and by proxy all the software is terrible and one of the reasons why all software
0:16:21 manufacturing is bad as a SaaS vendor is because the context you need from the customer is so deep
0:16:25 that almost every SaaS company builds the wrong product so the counter example for that is if I
0:16:30 gave anyone in this room a couple of software engineers and you interviewed 10 HR managers
0:16:35 and said build an onboarding workflow you’ve all been onboarded as employees you naturally get it
0:16:39 manufacturing has been such a hidden problem for such a long time that vertical integration for us
0:16:45 it just means that one it’s the only way to build the correct system and software because you’re
0:16:50 really feeling the right problem and pain and then the rapid iteration or one of our rules is
0:16:55 automators operate which is like you can’t build software for the factory until you can run that
0:16:59 part of the factory yourself there was literally the only way to build automation that works in the
0:17:04 real world versus a cool industrial robot that people buy and put on the shelf it’s a context problem
0:17:10 and you got to own your own pain I love that phrase own your own pain that’s 100 true it is
0:17:15 what Boeing used to do when they first invented autopilot for landings on the three last test
0:17:18 flights they just made all the chief engineers who built the autopilot software go on all the test
0:17:24 flights that’s the best bug fixing process in the world and that’s what you get with too many layers
0:17:28 removed it just doesn’t work for us it’s we need to be the fastest and highest quality contract
0:17:33 manufacturer and we look at what software what solutions are out there and we realize that the
0:17:39 design of wire harnesses is also incredibly fucked up and so when we get like excel spreadsheets and
0:17:43 PowerPoint slides from some of the largest companies out there and have to tell them this thing won’t
0:17:48 work yeah so you all touched on software there sometimes there’s this misconception that companies
0:17:53 like your guys’s are not software companies because you are primarily building physical
0:17:58 products and running physical processes I’d love to have you guys comment on that I don’t know
0:18:04 this is contrary anymore but selling sass software to hard tech companies is almost impossible it’s
0:18:09 very very difficult because it’s always going to be a unicorn solution I agree that way center makes
0:18:13 money is because we sell like a physical product that’s what the world needs more of versus some
0:18:16 of these other sass tools I mean the second thing that might be more contrary is like software is
0:18:20 not going to solve all your problems sometimes writing it on a post-it note and giving to the
0:18:24 first and next year it’s going to be a faster process than having all the software overhead I
0:18:29 think the way software helps manufacturer companies for us is to be scalable I call like the McDonald’s
0:18:34 approach which is if you want to work at McDonald’s within an hour you could start flipping burgers and
0:18:38 start building something and that just goes from just being able to like look at a screen and just
0:18:44 execute versus all the manufacturing companies I’ve worked at from large companies like Ford to
0:18:49 smaller companies like it is going to take you months and almost years to be just like really
0:18:52 skilled in a craft and actually know what the hell you’re doing and so I think that’s the whole
0:18:56 like plug-and-play unscalable model that in software is really really important I couldn’t have said
0:19:01 it better the reason why we have five software companies is because there’s two big problems
0:19:05 to solve in autonomous manufacturing one is the task layer which is someone takes 40 hours to do
0:19:10 something make it two hours and the second one is the coordination layer and we don’t believe in
0:19:15 full automation what we do all across the company is take a 40 hour task down to two hours and then
0:19:20 make the two hour task in software so that it’s de-skilled enough that we can train anyone in 30
0:19:24 days and that’s the only way to build fast enough and you also get a really resilient workforce the
0:19:29 other thing and probably why we’re going to win this market is every single deep tech company
0:19:34 in the history of trying to automate manufacturing has looked at four to ten big stages of really
0:19:39 expensive labour or coordination across a factory system whatever product you’re trying to build and
0:19:44 said this is too aggressive I’m going to build one of these bits first and then venture capital my way
0:19:49 and eventually I’ll build all five of them and it doesn’t work so one I think we had the right
0:19:55 culture but secondly you literally just have to build six products simultaneously at a ludicrously
0:19:59 aggressive pace otherwise you end up with a system where the front end is really really efficient
0:20:03 but the backside is really inefficient or you haven’t built that muscle and there are so many
0:20:08 problems in manufacturing that have real cost and schedule and quality ROI and by the way when I’m
0:20:12 talking about scarce labour I’m talking to like some of these roles we hired 20 experts from
0:20:17 manufacturing and then ran out of talent in the entirety of LA it’s that scarce but the big thing
0:20:22 for us of like full stackers we literally have to build everything one thing you touch on there is
0:20:28 the users of the software you develop are often employees of Patreon and there’s this dynamic
0:20:33 then because you’re doing everything vertically integrated you have in your case machinists
0:20:38 and software engineers and roboticists and so on all working in one organization can you talk
0:20:43 a little bit about bringing all those different disciplines together under one roof it’s extremely
0:20:49 complicated and an example is so there are I think five distinct cultures at Hadrian one is
0:20:55 Colet Silicon Valley software engineers our technician base which is going to be the
0:20:59 largest and most important part of our company over a long period of time kind of expert
0:21:04 manufacturers like very technical experts in machining or quality or whatever and then everyone
0:21:10 else and what you want to do as a leader of a company is say we’re all running in the same
0:21:13 direction right but often that has clashing incentives so if it comes out of the CEO’s
0:21:19 mouth like your job is to automate everything people in manufacturing have been burnt for
0:21:23 20 years by really bad bosses really bad leadership of you’re going to lose your job if I make you
0:21:28 more efficient I’m going to fire you so this is huge 30-year build up of oh I really hate everything
0:21:33 about this but we started from day one with culturally equal footing and then an economic
0:21:39 equal footing to the extent that California allows me to everyone is on a salary and if you are one
0:21:43 of those manufacturing people who are helping operate and develop software you have the same
0:21:47 equity bar as like a software engineer because we’re growing so fast like you’re never going to
0:21:52 lose your job unless you’re underperforming but we really had to build the kind of level set economics
0:21:56 and the culture from day one was and this is the biggest mistake private equity in Silicon Valley
0:22:02 makes the private equity version of this is you wear camo pants and I wear a suit so screw you
0:22:06 I know what you’re doing and the Silicon Valley version of that is I have a PhD I know what I’m
0:22:12 doing and both completely disrespect the core of manufacturing which is there’s nothing new
0:22:16 that needs to be invented here you’re just you got to pick 10 smart people and get the knowledge
0:22:20 out of their heads from 30 years of experience and get it into software and scale the hell out of it
0:22:24 we knew that coming in but it is a constant fight and every communication move you make
0:22:29 has to continually reinforce that and it comes down to really little things it’s not just the
0:22:35 executive of that operating division like machining or quality is giving product feedback
0:22:41 there is at least two to three literally hourly workers or experts that have to use that software
0:22:45 yelling at a software engineer being like your product isn’t good enough across the whole company
0:22:51 that does seem like a version of own your pain but I actually think that when you’re
0:22:56 putting in software or any kind of automation you can go too far too fast and just create more
0:23:00 process and work especially if you’re just taking something out of the box and shoo horn it into your
0:23:05 business and then productivity is down and everybody’s miserable one of the things that
0:23:10 we’re trying to do is actually look at from like a efficiency productivity aspect you bring in
0:23:16 automation where it has the most impact and then you fine tune that so it’s not painful
0:23:20 there’s a line that says if you want to get something done give it to a lazy person and so
0:23:24 I think I’m very happy I’m a lazy person because I didn’t get manufacturing you really have to do
0:23:27 the least amount of steps and the least amount of work to get something done
0:23:33 well off so not a deterring from the quality I think the culture et cetera is that our best
0:23:37 technicians the people actually building the stuff they’re not going to actually say what’s a
0:23:41 better process and you watch it in and out and you see someone just they’re cranking through
0:23:46 those potatoes and putting it on the fryer their whole goal is to not get fired and it goes fast
0:23:50 as possible it’s really on the manufacturing engineering team when I’m politely saying
0:23:55 we need to get this build from one week to one day what are all the inefficiencies because at
0:23:59 the end of the day people just want to put their head down from the labor force side and even at
0:24:04 SpaceX we did a time study and wire harnessing is a very labor intensive part there’s no machines
0:24:09 or tools to actually do it about 40% of the process of just a technician was just looking at a
0:24:12 computer and figuring out what the hell should I do when reading specifications and if you took
0:24:17 someone from a manufacturer and not from a data driven side you would say oh we should like
0:24:21 buy this machine and have it do this and that’s how I think software can really effectively
0:24:26 change any manufacturing flow there’s a word that’s come up a couple times now in this conversation
0:24:31 and that’s scale and I think intuitively people understand that with the kinds of processes
0:24:36 that you guys are building like a lot of the gains are fully realized at scale there’s different
0:24:40 models of this that you see around like scaling up versus scaling out like more modular approaches
0:24:47 and so on and so forth but what do senra, hadrian and castellan look like as manufacturing operations
0:24:51 at scale I believe for us we have a never-ending customer base and it always fits me as a founder
0:24:55 to say no to work because we don’t have the capacity and I think you have to scale with
0:25:01 machines automation and a cheap labor force ultimately in manufacturing right now people
0:25:05 ask me like what’s the best wire harness factory in the world and it honestly is not a contract
0:25:09 manufacturer it’s like a space x or a Boeing or someone else who vertically integrate into their
0:25:13 shop can pay people a little bit more because it goes into their end product and be able to
0:25:18 handhold the whole process we’re in a little bit different position because what scale looks like
0:25:24 for us is how many guided missiles will the us buy in a year it’s actually a published number
0:25:31 but it’s not as many as you would think a large order would be like a thousand a year and compared
0:25:36 to almost any other manufacturing industry that’s a joke of a number and so there are different
0:25:43 choices that will make because you don’t necessarily put in the expenditure to drive your manufacturing
0:25:49 system to the utmost best efficiency because the capital cost of implementing that would be so high
0:25:54 you would actually raise your cost of goods sold on you know a unit quantity of a thousand so we
0:25:59 actually have to try to look at where that balance is and I think where traditional aerospace
0:26:04 does this horribly is when they’re developing something new they make a bunch of decisions to
0:26:09 actually say well I’m only building one or two or five prototypes during this development and they
0:26:14 make a ton of design and manufacturing choices and then you get to the end of that and it works
0:26:20 and the government’s like okay I want to buy some they’re like whoa we can’t do 50 that’s crazy this
0:26:25 is impossible we only made this to build five and fundamentally they built the wrong product and so
0:26:28 we try to look at it from the other example but always holding the quantity of what we’re going
0:26:32 to build in mind as we’re designing it I think that’s absolutely critical but it’s also more
0:26:38 thoughtful than just saying I want to go for unlimited quantity and the hell the cost so
0:26:43 it’s an interesting balance. I would say everything that we build is built for scale not in the sense
0:26:48 of we’re building systems it’s just you got to get it as close to order comes in and order goes out
0:26:52 and the only problem is market share and growth and because there’s so much complexity in manufacturing
0:26:56 that is a tall order but this is why manufacturers don’t scale is because everything is a human’s
0:27:02 brain getting the job done so that’s how we think about it is every single error or process handover
0:27:06 or whatever slows the growth curve down because it just causes friction in the whole system.
0:27:09 I actually think that’s a great point how do you scale your engineering talent and how do
0:27:16 you make sure your thousands employee is as productive as your like 50th which if you’ve
0:27:21 noticed generally it starts to go down till you get to a certain company size and you can add
0:27:26 infinite people and you’ll probably get less work done how do you actually scale that that’s I think
0:27:31 one of the toughest aspects of running a business. We’re coming up on the tail end of our time here
0:27:36 and I want to bring it back to a very similar place to where we started around the big picture
0:27:42 of geopolitics and the future of the industrial base and how those two things fit together.
0:27:46 Brian you have on your website the phrase piece through deterrence and we talked a little bit
0:27:50 about how what Castellian is building is a non-nuclear deterrent but I would love to get
0:27:56 your thoughts on how our production capacity and how the manufacturing processes that you guys are
0:27:59 building that as a part of our industrial base how does that function as a deterrent.
0:28:05 Yeah so the stat that Chris said earlier about basically if we go to fight against China we’re
0:28:10 going to have weapons for three days and then we’re out are you scared of that? So when you look
0:28:15 at industrial capacity it’s not just a soundbite it’s literally what keeps the country safe and
0:28:20 keeps the ability for the country to do all these other things that have nothing to do with not
0:28:25 only industrial power or military superiority and I think to take care of the democracy you
0:28:30 have to actually have military strength economic strength cultural strength industrial strength
0:28:36 you have to have all of them we’re focused on actually enabling through an industrial renaissance
0:28:40 which is what I think a lot of the hard tech companies these days are actually trying to do
0:28:45 but that fundamentally keeps everything else in the economy humming along because if you don’t
0:28:50 have that like the US is on top for a number of reasons but not the least of which is that
0:28:54 a lot of countries would be stupid to mess with us but I think looking towards the future it’s
0:28:59 completely fixable we have everything we need to bring this back you just have to actually care
0:29:04 and focus on it and get back to the basics and I think the good news is I feel like people have
0:29:08 woken up to this is really a problem and they’re actually willing to invest time and energy and
0:29:14 effort to do something about it. Chris we spoke on the A16Z podcast about two years ago and we
0:29:19 discussed how fragile this ecosystem is and especially the average age of a machinist is 55
0:29:23 you know it’s probably higher now I would imagine and those are the ones that supply most of the
0:29:27 primes with their parts have there been any changes since then in the two years since and I know
0:29:30 that two years is a very short amount of time in the areas we’re talking about but have you observed
0:29:38 any kind of shift in the ecosystem? The vibe shift in state DOD in congress so two years ago everyone
0:29:43 was talking about procurement cycles and why can’t the government buy products at least over the last
0:29:48 12 months everyone is oh crap like we could buy all the products we want but there is no production
0:29:54 capacity across shipbuilding munitions vehicles whatever which is very good because at least
0:30:00 now we’re talking about what I think is the right problem on everything else honestly not really
0:30:06 to give you a statistic there is something like just in shipbuilding like just at Huntington Ingalls
0:30:11 and by the way China is building I think 20 to 50 times more ships than us every year
0:30:15 and to give you a personal story as an Australian who moved to five years ago
0:30:22 if the US Pacific fleet cannot defend Australia’s trade lanes the country runs out of food
0:30:28 in 27 days and runs out of oil and gas in 60 days to your point around resiliency it’s like
0:30:32 how rich is this country and how much do all of us in the room not have to deal with the
0:30:37 problems of bad guys or geopolitics and the answer is we have Coachella and Burning Man
0:30:44 it’s the most ridiculous safety abstraction of welfare hardistry whatever you want to call it
0:30:47 but anyway the macroeconomic picture gets worse every day because people get older and the
0:30:52 replacement rages isn’t anywhere near catching up the good news is I think over the last 12 months
0:30:57 there’s been a real vibe ship to oh shit it’s not just about buying products it’s about capability
0:31:02 capacity and the analogy I keep using is well 12 months ago we were talking about we need more
0:31:07 Netflixes and now everyone’s talking like oh we don’t have any data center capacity like at all
0:31:11 but you know I think there’s more people attacking the problem which is a really good thing but I
0:31:16 don’t think it’s going anywhere near fast enough there’s another phrase when you’re fed you’re not
0:31:22 hungry and I think this is why we don’t build things fast enough in the US and like why Skunkworks
0:31:27 Lockymar and built like the first fire jet in 143 days which is insane and I don’t think anyone
0:31:31 in this country can do that this fast enough but I think there’s going to be a wake up call in the
0:31:36 next few years especially when the labor shortage actually becomes more and more serious and we
0:31:41 might sound like the crazy people up here that’s like the sky is falling down but we’ve tapped
0:31:46 out the most skilled wire harness technicians in the LA area and it is really really difficult to
0:31:50 start scaling that but I think if we don’t do it now who’s going to do it 30 years later
0:31:57 all right that is all for today if you did make it this far first of all thank you we put a lot
0:32:01 of thought into each of these episodes whether it’s guests the calendar touchers the cycles with
0:32:07 our amazing editor Tommy until the music is just right so if you like what we’ve put together
0:32:13 consider dropping us a line at rapethispodcast.com/a16z and let us know what your favorite episode is
0:32:27 it’ll make my day and I’m sure Tommy’s too we’ll catch you on the flip side
0:32:28 you
0:32:28 you
0:32:29 you
0:32:39 [BLANK_AUDIO]

0:00:06 There are now drones here in the Bay Area that are flown from somebody that’s 20, 30, 50 miles away.
0:00:13 Mining has had some form of trucks without any drivers in them since 2007-2008.
0:00:17 China outnumbers our shipbuilding capacity about 200 to 1.
0:00:21 Immediately, every Autel drone was just bricked in Taiwan.
0:00:29 Basic assumptions about how software is built for the kind of traditional SaaS world of the 2010s
0:00:32 just doesn’t work in the autonomy space.
0:00:35 You can actually create a really, really incredible reconstruction of the world
0:00:37 just using these video generative models.
0:00:39 And this is not hype-gen AI stuff.
0:00:43 The cars we buy in US, in Europe, etc.
0:00:48 They’re not delightful consumer products like when you brought your first iPhone.
0:00:51 This has been a big year for autonomy.
0:00:56 For example, the fully autonomous Waymo driver has done over 20 million miles.
0:01:03 The equivalent of driving to the moon and back 40 times, and is now doing more than 100,000 rides per week.
0:01:05 But it’s not just autonomy on land.
0:01:12 For example, the FAA granted several operators the ability to fly commercial drones without visual observers earlier this year.
0:01:14 And this is only just the beginning.
0:01:22 Now, in this live recording from SF Tech Week, we brought in experts from three domains, Air, Land and Sea, to discuss autonomous systems.
0:01:26 And we touched on the real world deployments, the latest chips and their impact on the economics,
0:01:32 building full stack, quantifying risk, and regulations role in advancing this frontier.
0:01:37 Moderating this panel was a 16Z partner, Aaron Price Wright, along with three panelists.
0:01:42 First up, we have Makario Nemi, Chief Marketing Officer of Skydio.
0:01:45 Skydio is a 10-year-old company based here in San Mateo.
0:01:49 We provide quadcopter drones, more specifically camera drones.
0:01:51 We shipped about 50,000.
0:01:54 We started in consumer, especially for outdoor enthusiasts.
0:02:01 We’ve transitioned exclusively to selling into private enterprises, state and local governments, as well as the federal government.
0:02:08 Next, we had Vijay Patnaik, Head of Product and Applied Intuition, but also previously spent five years at Waymo,
0:02:11 most recently as Head of Product of their self-driving truck division.
0:02:18 Applied Intuition is focused on providing developer tools and software to companies that are building wake-ups.
0:02:24 We provide simulation and data tools that are necessary for building autonomous systems.
0:02:31 And finally, Peter Bowman Davis, an engineering fellow now at A16Z, but previously worked on machine learning at Soronic.
0:02:34 Soronic is a full-stack maritime autonomy company.
0:02:38 We build boats that are autonomous and we sell directly to the government,
0:02:44 so that comprises everything from the whole manufacturing to the simulation side on the autonomy.
0:02:47 We also had some amazing questions from our live audience.
0:02:49 So stay tuned for that at the end.
0:02:53 And of course, if you’d like to attend events just like this in the future, make sure you’re subscribed.
0:02:56 And while you’re at it, leave a review.
0:02:58 All right, let’s get started.
0:03:03 As a reminder, the content here is for informational purposes only.
0:03:06 Should not be taken as legal, business, tax or investment advice,
0:03:13 or be used to evaluate any investment or security and is not directed at any investors or potential investors in any A16Z fund.
0:03:19 Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast.
0:03:25 For more details, including a link to our investments, please see a16z.com/disclosures.
0:03:35 So I’m really excited because we have someone from the business, from the product,
0:03:39 and from the tech side here, as well as Air, Land and Sea.
0:03:48 So maybe to get started, I would love to hear how far along in the journey to autonomy you feel like we are in your industry,
0:03:54 maybe framing the levels of autonomy and where are we really all the way there and where do we have still some way to go.
0:04:00 So Skydio was sort of started on the foundation of autonomy and we did not invent the quadcopter.
0:04:02 Those existed prior to the founding of the business.
0:04:07 I think we’ve all experienced or seen them and even just walking around Best Buy sort of toy form,
0:04:12 where you have some sort of radio controlled mechanism to be able to tell the drone where to fly.
0:04:18 The challenge with the use of drones is that it always required a person to be there.
0:04:26 And if the more you go up in the use cases into more enterprise or government use cases, the more valuable the flying becomes.
0:04:33 And so if you’re going to go fly around a substation that belongs to an energy utility and you go crash that drone,
0:04:35 you could take out power into an entire neighborhood.
0:04:39 That’s not just something that’s cute that your kids did on Christmas afternoon.
0:04:43 That’s actually something that would cause you to lose your job because real problems for the community.
0:04:50 So the necessity for the person is not just to be there, but they have to be an expert pilot.
0:04:54 They have to be great at being able to fly in these really challenging environments.
0:04:56 And that was the premise of Skydio’s.
0:05:02 Can we build the skills of an expert pilot into the drone itself so that any one of us here in this room can actually pick up the drone.
0:05:07 It would be reasonably proficient and they become sort of democratized in terms of their access.
0:05:15 So we started day one from building an autonomy and the first iteration of that was the ability to follow somebody.
0:05:24 Since then, we’ve sort of upped our game quite a bit, not only beyond obstacle avoidance, but to be able to not have a person be present at all.
0:05:33 So going back to that substation example, there are now drones here in the Bay Area that are flown from somebody that’s 20, 30, 50 miles away.
0:05:36 And those drones complete automated inspection missions.
0:05:37 So we’re getting there.
0:05:43 The regulatory environment doesn’t allow us to just completely eliminate the person, but we’re certainly making quite a bit of progress.
0:05:53 In the automotive space, which is passenger cars at the lower levels of automation, what we would think of as for a user hands on system where you’re still paying attention to the vehicle.
0:06:00 Your hands are on the steering wheel, but the vehicle can do things like automatic emergency braking or what’s called cruise control.
0:06:05 Those systems I think are available on sort of any car you go and buy today in Europe.
0:06:08 Some of those are mandatory from a regulation standpoint to be there.
0:06:22 I think US regulations are slightly behind in mandating some of those systems, but generally those are available on vehicles where the next focus area for a lot of the OEMs is on what’s called sort of level two plus systems, which are these hands off system.
0:06:24 So as a user, you don’t need to keep your hands on the steering wheel.
0:06:27 You still need to pay attention and keep your eyes on the road.
0:06:30 And one level beyond that, but it’s eyes off as well.
0:06:32 So you can read a book, you can watch a movie.
0:06:38 Those systems aren’t yet deployed. That’s in R&D phase right now, just very early deployments right now.
0:06:46 And then on that level four side, I think in the city in San Francisco, we’ve all seen sort of Vemos robot axes and that’s the sort of main deployment on the car side right now.
0:06:50 And I think there are some deployments in China that we see on the level four side as well.
0:06:53 Taking a different industry, I think maybe construction and mining side.
0:07:01 What’s interesting is mining has had some form of trucks without any drivers in them since 2007, 2008.
0:07:06 It’s not the same level of autonomy as Vemo in the sense that those trucks are following up predefined paths.
0:07:10 They get a lot of support from the infrastructure in the mine in order to do that.
0:07:13 But there’s a clear business case and an ROI for that product.
0:07:20 And hence the mining industry has been investing in that the technology itself is far behind what’s available on today in automotive.
0:07:24 So there is a big focus to upgrading that technology and serving more of the use cases.
0:07:28 Maritime I would say takes a lot of inspiration from the self driving car community.
0:07:35 And the reason for that is effectively when maritime you have this kind of two dimensional plane that you’re moving around very similar to self driving cars, right?
0:07:36 You just have lateral planning.
0:07:40 It’s maybe a little bit more complex because you have the longitudinal axis as well.
0:07:43 And you also have much less things in your environment to ground you.
0:07:48 So when you’re thinking about like an autonomy model, the world doesn’t usually just like nicely translate around you.
0:07:53 Think about driving a car in a nice kind of like city skyscraper block.
0:07:56 As you take a left, the world nicely transforms around you.
0:08:00 But imagine you’re in the middle of the ocean and you take a left.
0:08:03 The ocean looks the exact same.
0:08:10 And so you actually get a lot less information per frame when you’re making these sort of like vision models or you’re making a sort of autonomy stack.
0:08:16 And then the other thing that I wanted to draw is the delineation between a perception stack and an autonomy stack.
0:08:20 For a lot of these use cases for us at least perception has mostly been solved.
0:08:23 That is to say object detection, object avoidance.
0:08:30 These are like very, very simple tasks and they can mostly be done with CNNs, which are 20, 30 years old.
0:08:36 Autonomy is a bit trickier because you actually have to take kind of elements from the perception stack and you have to make them actionable.
0:08:39 And that is to say you have to do a lot of long term planning.
0:08:42 You have to actually take in that data and make decisions based on it.
0:08:45 And that’s something I think that’s only been enabled in the last five.
0:08:47 Maybe you can make the argument even two years.
0:08:52 And so I’m super excited about a lot of the work that’s going into reinforcement learning and long term planning.
0:08:58 Because this is super, super important in the maritime domain because in maritime, you’re not often near the shore.
0:09:03 And in the defense application, you’re not often in communications with back home.
0:09:05 And so you need to make a lot of independent decisions.
0:09:09 And so these are the things that we’re thinking about in the maritime autonomy space.
0:09:12 Double clicking on some of those sort of technical breakthroughs.
0:09:17 Where do you think the latest developments in AI have impacted autonomy?
0:09:24 To what extent have they impacted autonomy and how much change are we seeing in the industry today versus let’s say five or 10 years ago?
0:09:32 Just based on the current state of the art and things like video language models or other new transformer based architectures that might not have existed.
0:09:36 For example, when Waymo got started or when Skydio got started.
0:09:41 Yeah, I think pretty significant impacts actually in terms of the latest developments.
0:09:46 I think foundation models do have an impact on the architecture of these autonomy systems on these ground based systems.
0:09:52 Where you think of foundation models as replacing a number of more task specific models that existed.
0:09:59 So we see more and more companies making the shift towards using or at least researching with how foundation models can be used.
0:10:04 The quote unquote frontier of research right now is end to end.
0:10:11 Even when I joined Waymo back in like 2016, there was a lot of hype around end to end driving, but it didn’t really materialize the technology wasn’t ready.
0:10:17 So now it’s almost a second version of that happening now where the research has progressed a little bit further.
0:10:23 So I do think the architectures are going to evolve significantly even though they’re not yet ready for production.
0:10:28 What that also means is that the tools and infrastructure needed to support that is evolving pretty quickly.
0:10:35 And so we are developing new kinds of simulators that are needed in order to support those advanced architectures.
0:10:43 And those simulators themselves require many of these generative techniques or neural rendering techniques that have shown good promise in research.
0:10:48 And third, I would say is just using some of the generative AI to simplify workflows.
0:10:56 As an example, if you’re not on the engineer, you spend a lot of time working with simulation and scenarios, but you don’t want to write a lot of complicated scenarios.
0:11:04 But we could generate them for you programmatically in a much more confident manner today than what was possible two or three years ago.
0:11:10 So just at different layers, whether it’s the autonomy stack, whether it’s the tools being used to develop that autonomy stack.
0:11:15 I think we see pretty fundamental impacts and that’s why we are investing pretty aggressively in utilizing these technologies.
0:11:19 Just double clicking on the idea of simulation for robotics or autonomy.
0:11:24 I think this is like a very non obvious point to people who haven’t worked in this industry for the last two years.
0:11:34 Because I think that we’ve started to use video generative models as dropping replacements for simulators like Unity or Ross or Unreal Engine.
0:11:42 And the reason you need a simulator to be clear is because, first of all, robots are really, really expensive and you don’t want to take them out to the field to have them break if you mess up the algorithm, basically.
0:11:51 And you also need many environments where you need to train on policy, meaning you basically need to run it in real time in order to make sure the thing works.
0:11:57 And so in recent years, they’ve started to use video generative models conditioned on actions or sensor data.
0:12:02 And you can actually create a really, really incredible reconstruction of the world just using these video generative models.
0:12:08 And this is not hype Gen AI stuff. This is used by Tesla, this is used by Wave, it’s pretty exciting stuff.
0:12:12 And I think that some people are being very quiet about it, but there’s some really awesome public releases.
0:12:25 I will say the rise of NVIDIA has been pretty instrumental for our kind of business to be able to put that kind of compute power in a four and a half pound form factor that can be sold for basically $11,000, $12,000 would still make some margin.
0:12:27 And that’s a very hard thing to do.
0:12:33 I’m curious, how do you think about the cost aspect when using the high performance chips that NVIDIA comes out with?
0:12:39 How does the sort of ROI for the business change when you’re putting the latest GPUs or compute onto the drones?
0:12:45 It’s so foundational to our business. We have to have high computes on the drones themselves.
0:12:51 What we did in our latest release that we announced about a year ago was we actually added excess capacity.
0:12:56 So basically trying to feature proof the hardware so we can do more over time. We haven’t fully used it.
0:13:03 We expect to be able to run additional models, including models that our customers build on the drone itself because they can do much deeper analysis on the things that matter to them.
0:13:07 What’s an example of a custom model that a customer might want to build?
0:13:14 So object detection is a really simple one, as you mentioned, but in our world we see everything.
0:13:21 It’s not just about, hey, people, right? It’s like, no, I want to be able to identify that particular transformer.
0:13:26 And by the way, PG&E’s transformers are different than Southern California Edison’s transformers, which are different than Baltimore Gas and Electric Transformers.
0:13:34 So they know their transformers and then once we actually identify the transformer, we need to be able to determine whether the thermal signature is actually saying there’s a problem or not.
0:13:44 That is so specific. We’re not the best ones to be able to build that. So that would be an example of how customers can build their own stuff and then basically run it so we can take immediate action on the drone itself.
0:13:50 That’s why we put in that excess capacity is to be able to do those kinds of things that have that kind of extensibility and platform.
0:14:01 Very cool. Actually, that brings me to another question, which is for Ceronic and Skydio, both companies are essentially building the full vertical stack where not just the algorithms around autonomy and perception,
0:14:09 but actually the full kind of product that goes into the world. Whereas Applied, you’re much more of a software provider working with companies who are building the hardware.
0:14:21 I’d be interested to hear some of the kind of trade-offs or what do you gain by owning the full end-to-end system and what’s harder as a result of having to manufacture and deliver the final product.
0:14:28 We think that the ultimate value of a vertically integrated stack is reliability. We’re very committed to it. We started with it.
0:14:36 And in fact, our primary competition in the world is DJI, the largest manufacturer of drones out there, and they are the complete opposite.
0:14:42 They’re happy to build the hardware and there’s a whole plethora of other companies to sort of build their ecosystem around DJI hardware.
0:14:48 And we see that those seams actually start to fracture when you get more and more complicated scenarios.
0:14:55 And so you add in autonomy there. What happens in the corner cases? What happens in the failure modes? How do things react?
0:15:00 That’s where we could completely control that and we can test for it and we build reliability around it.
0:15:07 So I think the downside is that we might go slightly slower, right? Because we do have finite resources.
0:15:16 We have to choose where our investments are. Whereas another company that may be all of what they do is to be able to build some kind of capability on top of another person’s hardware.
0:15:21 For us, we are okay to move a little slower to be able to deliver higher quality.
0:15:25 We’re kind of like a weird middle ground, I would say, because we own the full stack.
0:15:29 We also, though, have three types of boats, like small, medium and large, basically.
0:15:34 And these boats are very different in terms of the amount of cameras they have, the actuation techniques you use.
0:15:37 That being said, we still own all the hardware, especially all the compute.
0:15:44 And so the compute is mostly homogenous between them, but your actual kind of effector systems or your perception stack is going to be differentiated.
0:15:49 And it’s also differentiated in between boats sometimes, because we want to test things out, we want to add another camera or something.
0:15:54 And I would say it’s actually a powerful thing to be a little bit agnostic on the hardware and the software,
0:16:02 because it basically pushes you to create really, really solid abstractions for your perception stack or for your autonomy stack.
0:16:06 And it kind of pushes you to be a better engineer. This is at least what I’ve experienced in my engineering team.
0:16:10 We really liked to think about, okay, how is this going to work on boat N+1?
0:16:15 And so that’s been kind of a fun challenge to work on, but I do agree that it does slow things down sometimes.
0:16:18 Yeah, I completely agree with that point of abstraction.
0:16:21 So Applied started with being a tools provider.
0:16:26 So we would provide the simulation and data products and eventually our customers were like the tools are great,
0:16:30 but we don’t have the internal capability to build that autonomy stack using the tools.
0:16:31 Can you help us with that?
0:16:33 So that’s how Applied actually started.
0:16:35 We now provide an off-road autonomy stack.
0:16:42 We also have a trucking stack and to the point that Peter was making now more so than ever in autonomy.
0:16:48 I think these abstractions are possible where we have a stack that it’s not built for a single customer.
0:16:53 It’s meant to be reusable across customers and they can build their eventual application on top of it.
0:16:55 So if you only want to use the tools, that’s great.
0:17:01 But if you want an actual autonomy stack, either as your primary stack or to get kickstart your efforts internally,
0:17:09 there are providers like Applied or other ones that you can use with the right abstraction to significantly accelerate your program.
0:17:16 How involved is that sort of translation process to different form factors, vehicles, customer types?
0:17:22 I’m curious to hear how hard it is to get autonomy that works for one type of vehicle to work in a totally different setup.
0:17:26 I mean, I would love to tell you that it’s like seamless and just happens all by itself,
0:17:30 but there’s the engineering aspect of it and there’s the organizational aspect of it.
0:17:34 I think engineering to a certain extent, you can engineer the best APIs and abstractions.
0:17:41 There’s still customization that you would need to do, especially if it’s for the first time being deployed to different platforms.
0:17:48 But the organizational challenge is even more interesting because for some of the companies that we are working with,
0:17:56 they’re going through this big shift of going from hardware companies because vehicles were primarily hardware driven to becoming software companies.
0:18:03 So they’re reinventing themselves, hiring software engineers, hiring engineering leaders from Silicon Valley, etc.
0:18:09 And you actually need a very close partnership to make this work, especially if there’s various different vehicle form factors.
0:18:17 So it’s not just an engineering problem, it’s also how do you build up internal capabilities for customers who often will work with them
0:18:24 on like training their internal team, etc. in order to be able to work together on this transition of the stack to many different platforms.
0:18:30 Makes sense. I spent most of my career before investing in Palantir, so I’m very comfortable with that type of model.
0:18:37 What was your experience there in terms of how to help companies sort of make that shift to whether it’s autonomy, whether it’s software, data?
0:18:41 It wasn’t necessarily strictly autonomy, but I think that there are a lot of similarities.
0:18:45 The biggest thing was how you embed with the customer to drive cultural change,
0:18:54 which is the Ford-deployed engineer model really invented at Palantir where I would be flying out to Azerbaijan or somewhere in Oman or Trinidad
0:19:00 or really far-flung places to help their engineers really on the ground in the field figure out how to use the software.
0:19:02 And so maybe switching gears a little bit.
0:19:08 I’d love to talk a bit about the economics of autonomous systems, also how that relates to regulation.
0:19:14 So we’ve heard and read in the news a little bit lately about Waymo and the economics of a Waymo ride.
0:19:21 And it’s really exciting to see it start to actually make sense so that you’re not just pouring money into the system for the sake of R&D.
0:19:30 But every single Waymo ride is actually, I think, profitable if you don’t count the overall cost of the vehicle, which is a big cost to discount.
0:19:37 So I’d love to hear maybe starting with Skydio, like how do you think about the economics of the drone industry, where you’ve seen real success,
0:19:41 where it’s been slower, and maybe also how regulation has played a role in that.
0:19:46 We serve multiple different industries and their ROI calculations are going to be slightly different.
0:19:49 I’ve used a couple of examples where we talk about energy utilities.
0:19:58 And in that instance, it’s about looking at the people and labor it takes to go inspect the equipment, whether that’s physical substations or transmission lines or distribution lines.
0:20:04 And anytime you can save the cost of actually sending a truck on site where they can put a person in a bucket, get a buck in the air,
0:20:10 which can cost $2,000 per deployment, and you can instead put a drone up and get the exact same information in 60 seconds.
0:20:13 Just sort of a pretty obvious ROI there.
0:20:16 We also serve public safety, like first responders, police departments.
0:20:27 And I think you can’t really put a value on human life, but you can put a value on the insurance payouts that come with officer involved shootings.
0:20:32 It can easily go to $1 to $2 million every time there’s some kind of use of force.
0:20:36 And so being able to avoid that, being able to cut that down in half is really, really key.
0:20:40 Ultimately, we’re in the business of providing information for people to make better decisions.
0:20:50 And if you’re in a really high stakes scenario, better information about where the threat is can really make a big difference in terms of whether you end in a peaceful way or a tragic way.
0:20:56 Certainly some elements of economics, but a lot of it is just keeping people safe, keeping their officers safe, keeping the community safe across the board.
0:21:03 And how about with the regulatory piece of sort of whether an operator has a line of sight on a drone?
0:21:11 How maybe take a particular use case like police officers or something, like how they think about kind of fleet expansion as those regulations change?
0:21:19 One of the key areas of investments right now is a concept called drone as first responder, where you would pre-position docking stations on the rooftops of cities.
0:21:33 And so this is happening in New York, it’s being tested in other cities where when a 911 call comes in, the drone automatically deploys and goes and is effectively the first responder, the first person on site and is providing eyes back to a remote pilot.
0:21:34 That does two things.
0:21:43 One is the pilot is sitting in some nicer conditioned office and they can immediately relay information back to any responding officers so they can respond more appropriately.
0:21:49 They can control many, many drones at the same time, so it becomes a one to many versus a one to one argument.
0:21:51 And a lot of calls can be cleared.
0:21:56 These 911 calls for service can be cleared by basically the drone saying there’s no real issue here.
0:21:59 You don’t have to go out on site and that’s tremendous economic savings.
0:22:08 In order for this drone as first responder to happen, you have to have a regulatory environment that allows for a lack of visual observer on the rooftop.
0:22:14 And right now, just the 22nd primer on FAA regulations, if you’re putting anything in the air, it is regulated.
0:22:17 The FAA cares about airspace and they mostly care about just not hitting anything.
0:22:19 Don’t hit an airplane, don’t hit a helicopter.
0:22:26 And so they require you to stay below 400 feet and they require a person to be visually looking at that drone at all times.
0:22:27 That’s like the most basic rules.
0:22:31 So now you have a scenario where you might go further up.
0:22:34 You don’t have the available staff to put a person on the rooftop to be able to look out.
0:22:40 And the second is if you’re in some place like New York City, the top 100 buildings are over 600 feet.
0:22:42 So how are you going to be keeping everything below 400 feet?
0:22:50 So being able to sort of work with the FAA to be able to create these kinds of waivers to this traditional regulatory environment that allows for beyond visual line of sight,
0:22:59 beyond visual line of sight without a visual observer and to be able to actually stay within 50 feet of structure so you can actually go up and over a building as long as you’re within 50 feet.
0:23:02 And that was a waiver that just came out actually two, three weeks ago in New York City.
0:23:09 And it’s a fantastic way of having the FAA sort of work with these agencies to do us right for the community while still keeping the air safe.
0:23:11 Exciting. Congrats.
0:23:13 I’d love to hear the applied intuition.
0:23:23 The economic case where you see where is this still sort of a slightly money losing R&D exercise versus where are you actually starting to see the economics make sense?
0:23:26 And where, if at all, regulation plays a role in that?
0:23:36 Yeah, I mean, on passenger cars, they’ve had driver assistance systems for a long time as these systems become more capable OEMs have an ability to charge more for them, right?
0:23:39 That’s why the FSD system from Tesla when it came out was over $10,000.
0:23:42 There’s an Avid Tesla fossil driving user.
0:23:43 There you go.
0:23:44 I get a test.
0:23:45 Yeah.
0:23:48 So there’s a certain willingness to pay if the performance of the system is up.
0:23:58 They’re not just discounted that. So I think the industry is in a price finding mech sort of stage right now where obviously the best thing would be higher performance while keeping costs the same.
0:24:04 But as the cost of compute goes up, more sensors go into the vehicle, more software being deployed onto it because they are more complex.
0:24:06 The bomb cost for the OEM goes up.
0:24:08 So I think there’ll be some price increases.
0:24:13 OEMs already operate on somewhat thin margin compared to the margins we are used to in the Silicon Valley.
0:24:16 So they’re not in a position to keep losing money on each of these systems.
0:24:22 So I think the market will eventually find a price that makes it profitable for the OEM to ship these systems.
0:24:27 On the, I think areas like trucks and construction mining have a different unit economics framework.
0:24:35 So if you think about trucks, for example, and I’ll tie in regulation into this as well, we can think of unit economics as reducing the current costs.
0:24:39 But a big part of unit economics also can you increase the revenue per truck or not?
0:24:41 So the first argument is, hey, there’s a shortage.
0:24:48 So we’re losing like supply chain flexibility, like we do some work in Japan in providing our autonomous truck technology in Japan.
0:24:58 The government is actually pushing the commercial vehicle sector to invest more in autonomy because they’re seeing they call it the 2024 problem where the drivers are aging out and they are overworked and have health issues.
0:25:00 So the shortage problem is real.
0:25:05 But even if you look beyond the sort of, okay, how do we reduce the labor cost and the insurance cost?
0:25:14 If you have a network of autonomous trucks, you can actually optimize the entire logistics network such that you’re generating more revenue per truck, which makes unit economics better.
0:25:17 The regulation plays a part in that is twofold.
0:25:23 One is in a place like the US, each state can actually regulate how these driverless trucks are deployed.
0:25:31 And so you can have this weird mix where the states that your truck is traveling through have slightly different regulations and federal government has different regulations.
0:25:33 So you do need some consistency.
0:25:42 So if you want to have a truck grow from LA to Atlanta, which is actually a pretty heavy freight route, then California has to have consistent laws with Arizona and Texas that allows that.
0:25:44 And I think it’s not fully there yet, but getting there.
0:25:51 The other side of it, there are regulations like hours of service that determine how long a truck and a truck driver can drive.
0:25:54 That’s for their safety from fatigue, et cetera, and for road safety.
0:26:07 So the government has to be willing to be flexible on those such that those trucks can then operate 24 hours such that you can then reconfigure logistics networks and then make the supply chain more efficient and make the revenue and the unit economics work.
0:26:16 So I think those are some of the factors in trucking and mining, I think is a different sort of set of factors in the sense that the ROI is somewhat already clear to the industry.
0:26:23 That’s why they’ve been investing in autonomy for a long time and the trucks we’re talking about in mining on which these autonomy systems are being deployed.
0:26:25 These are like a $5 million machine, right?
0:26:29 So it’s not a big deal if you put a $100,000 lidar on it.
0:26:35 Of course, it adds $100,000 and that’s meaningful, but in relative to a $5 million machine, that’s not the main point.
0:26:46 The main point is if the system stops and it stops the mining operations for every minute you stop the mine, you’re losing them tens and hundreds of thousands of dollars.
0:26:51 So the unit economics game there is at a mine level, not necessarily at a truck level.
0:26:58 I also imagine replacement cycles are different. It’s not as easy to buy a new $5 million bulldozer as it is to buy a new $100,000 car.
0:27:05 That’s right, but the whole ROI calculation becomes a slightly different calculation, which is how many trucks do you deploy in a mine?
0:27:10 How do they impact the productivity, which increases the mine revenue and how do they impact the downtime, etc.
0:27:18 So we see slightly different applications, but I think generally in mining, as I said, OEMs and the mines, we talk to see the business case there.
0:27:23 It’s just improving the technologies so that it can reliably work and not bring downtime to the mines.
0:27:33 So I think related to some of the questions around regulation, I’d love to talk a little bit about geopolitics and some of the regional or global differences in our approach to autonomy.
0:27:40 I don’t know if this is a true story, but I did hear the story about DJI drones being used early on in Ukraine.
0:27:54 China had given Russians a backdoor to be able to determine the positions of Ukrainians on the ground using DJI drones, which was a big impetus for a lot of the drone activity that we’re seeing in Ukraine coming from the U.S.
0:27:58 So there’s a lot going on in this kind of geopolitical landscape.
0:28:13 And I kind of love to hear how, for all three of you actually, given Sironic as a defense contractor, I’d love to hear how kind of the role of geopolitics and how you think about autonomy and how it’s come up both as maybe a challenge and also a motivating factor.
0:28:19 So Sironic was basically founded on this idea that China outnumbers our shipbuilding capacity about 200 to one.
0:28:36 So the question is, what can we use as a sort of unfair advantage to leapfrog that rather than just compete on the basis of cost per ship? How can we make autonomous distributed systems deliver sort of an overmatch result over China’s kind of shipbuilding capacity?
0:28:49 And so this is obviously played a major role in motivating Sironic’s accompany. But sort of even beyond that, I think we’ve seen a lot of unmanned service vehicles, which is autonomous boats, be used in the Middle East as well as with Ukraine and Russia.
0:29:03 And so I think the sort of heating up of global proxy wars, not just great power conflict, has also led to a lot of expansion of use of autonomous systems as we see in Ukraine, as we see off the coast of Yemen and other countries like that.
0:29:05 But yeah, that’s the TLDR.
0:29:09 Given the sort of DJI angle here, we’d love to hear how Skydio thinks about this.
0:29:16 Geopolitics is a very important topic in our world. You mentioned Ukraine. We’ve been in Ukraine 30 times. I was there this past spring.
0:29:25 They use DJI’s in a very disposable fashion because it’s the cheapest they can get and they hack it. DJI’s effectively a hostile vendor to them.
0:29:31 And so they have to hack it quite extensively so that the drones can actually fly there and not be detected by the Russians.
0:29:45 And there’s certain techniques they have to do that. One of the challenges that the U.S. looks at is the threat of being wholly dependent on Chinese based technology for industries that they would consider to be critical to national security.
0:29:51 What Ukraine has shown to the world is how critical drones can be in any combat.
0:29:58 And if all of our drones and all of our drone parts are being held by Chinese manufacturers, then we are wholly dependent on them for such a critical technology.
0:30:02 And that is exactly what the U.S. is trying to avoid.
0:30:10 And so there’s laws being discussed within Congress now to start weaning the United States off of Chinese manufactured drones.
0:30:19 So it won’t be an immediate ban. It won’t happen overnight, but they can start sort of lessening their dependence and start encouraging U.S. and U.S. innovators to step into the fold, including ourselves.
0:30:33 We’re not asking for subsidies, but we do benefit from restrictions for sure that for some of these organizations like the federal government or state and local governments to choose western products instead of Chinese manufactured products.
0:30:41 And one small example of the kind of power that some of these manufacturers can have, Otel, which is sort of a smaller version of DJI based in the exact same city in China.
0:30:48 They decided one day that they would update their geofence, basically where you’re allowed to fly or not allowed to fly.
0:30:50 And they turned off all access in Taiwan.
0:30:54 So immediately every Otel drone was just bricked in Taiwan.
0:30:56 Now, why Taiwan?
0:31:07 So if you had an example where all of our critical industries are using a Chinese manufactured drone, and one day China says, I’m just going to turn them all off, they could immediately break them.
0:31:10 And that’s really the extent of the problem. It’s quite severe.
0:31:19 Yeah, I think China has a conversation with every CTO or CEO of any major OEM across industry that visits our office or we talk to.
0:31:26 And these are like weekly conversations just to take an example from automotive and maybe government because applied also does a bunch of work with the government.
0:31:30 In automotive, China is basically redefining the industry.
0:31:32 There’s no sort of lighter way to put it.
0:31:36 There’s a few things that are happening and the Beijing Auto Show was earlier this year.
0:31:38 It happened to be there in person for it.
0:31:40 All of the attention was on the local Chinese OEMs.
0:31:48 Like you could literally go to the booths of the international companies and consumers didn’t really have much interest in them except for a couple of brands.
0:31:54 And that’s because the product innovation in the China ecosystem is second to none right now.
0:32:04 The cars you can buy there, like buy, not just drive in a prototype, just vastly superior consumer experience compared to what you can get in the U.S. or Europe.
0:32:12 I was talking to Steph earlier about some of the experiences there in terms of you walk up to a car and you say, please unpark yourself, it unparts itself, opens a door for you.
0:32:17 You sit in the car, the car assistant can process commands from four people talking simultaneously.
0:32:21 It’s a delightful experience to sit in those. These are production cars.
0:32:23 And it’s not just about the product today.
0:32:27 It’s actually scary how fast the pace of innovation is every six months.
0:32:32 They’re innovating on that product and the cost at which they’re able to do it is very hard for the global industry to match.
0:32:38 So at one point you might think, well, there’s enough geopolitics that maybe the global economy is like somewhat insulated from that.
0:32:42 And we’re seeing that in the U.S. where there’s 100% tariffs on Chinese vehicles.
0:32:50 At the same time, you look at the sort of what’s happening in the market in China where these OEMs are facing a sort of flattening domestic demand.
0:32:51 They have oversupply.
0:32:58 So even though China is the largest automotive market already and the largest exporter, they all have ambitions of sort of going global, right?
0:33:07 And some of these OEMs have cost points that are so low like a BYD that despite tariffs in Europe and despite tariffs in U.S., they might actually be somewhat competitive.
0:33:17 And that’s why every automotive executive is worried about China and thinking about what’s our strategy, not just in China, but what’s our global strategy given what we are seeing in China.
0:33:24 I think on the government side, it’s similar honestly in the sense that the DoD used to be the driver of innovation, right?
0:33:32 There’s a lot of ties between the Silicon Valley history and DARPA and how autonomy came about and how a number of other technologies like even internet came about.
0:33:42 And now we’re at a point where I think the Pentagon recognizes that it’s been slow in moving towards software and moving towards autonomy and definitely slower than China.
0:33:47 Both in terms of sort of processes of like how do you do procurement is actually a big barrier today.
0:34:01 So we do this conference every year in DC focused on national security and the entire conversation is about how do we move faster and bring these technologies that are being deployed in the commercial world to national security much faster.
0:34:04 And how do we evolve our procurement processes to be able to do that?
0:34:12 So I think across industries, we see China being a major part of conversion, a major driver of strategy for companies.
0:34:17 I was going to end with the last question of what keeps you motivated, but I feel like that’s very motivating.
0:34:25 So with that, I think we have a few minutes left, so I wanted to just open it up if anyone in the room had any questions for our wonderful guests.
0:34:33 Alright, since this was a live recording and some of the questions were pretty long, we’ve condensed them for your ears and I’m going to be punching in with the voiceover.
0:34:39 Alright, first question. What are your experiences dealing with security guidelines and regulatory bodies?
0:34:48 The one thing that I would say is it’s a little bit unintuitive sometimes that designing autonomous systems, as long as they’re not like true kinetic effectors, sometimes it’s actually lower risk.
0:34:53 The example that I like to give a lot is imagine designing a drone versus a helicopter, right?
0:34:58 Helicopter, you’ve got to be pretty sure that that thing will not fall out of the sky, right?
0:35:02 A drone like falls out of the sky. Okay, that’s fine. We’ll iterate, we’ll keep going.
0:35:08 And so sometimes it actually helps me sleep a little bit better at night thinking, okay, at the end of the day, these systems are directing themselves.
0:35:12 If the boat sinks, it’s not the end of the world. And so that’s what all adds to it.
0:35:22 In terms of the actual reality, the fact that we’re designing for the DoD though, yes, these systems have to be absolutely like secure mission critical, especially in terms of reliability.
0:35:26 And to this extent, I think CERONIC as an engineering org is built up around this.
0:35:33 Like the way we do our systems engineering, the way we do our test and evaluation pipelines, I think is a lot more rigorous than a lot of other companies.
0:35:38 But at the end of the day, you can sleep a little bit nicer just knowing that they’re unmanned systems.
0:35:44 From a product design standpoint, in terms of safety, a lot goes into basically failure mode.
0:35:50 So what happens if the battery gets too low, no matter what inputs the operator gives, the drone is going to come back.
0:35:57 And it’s going to follow the path that it had so it can sort of navigate back from obstacles, no matter where it’s out to get back to its original location.
0:36:01 It can also basically land directly down. There’s a programming of safe landing zone.
0:36:07 So those are all things to try and keep people sort of safe in terms of more security, especially with the Department of Defense.
0:36:13 We sell a variant of our products that are effectively offline and offline is a bit of a misnomer.
0:36:16 They’re online, but they’re completely within the private network of the DoD.
0:36:23 So we have no access to it. It’s not cloud based. And those are instances where if the customer really, really needs that level of security, they have it.
0:36:29 And it basically means more engineering work for us because we have to carry two variants forward, but it’s necessary in order to circle federal government.
0:36:35 Yeah, generally, in sort of the automotive realm and sort of Vamo cars, etc.
0:36:45 What I would say is that what’s been done in the industry in the past, like all of the systems engineering practices that have been used to build aircraft like commercial airliners, etc.
0:36:55 Are necessary and are used, but not sufficient in the same way in some of the standards that have been put out ISO 26262 that has been used in the industry for a long time.
0:37:01 These are all inputs, but none of these individually or even cumulatively are sufficient.
0:37:07 And so if you look at some of the publications from Vamo and some of the work we do with our customers, we actually consistently tell our customers.
0:37:14 We actually need to do way more than what’s actually any regulatory bodies asking for or what aerospace engineering has done.
0:37:29 In order to prove the safety of the system that actually validating that these systems are safe to be deployed, whether that’s a passenger car, whether that’s a truck, whether it’s a mining agriculture defense application is actually one of the problems our customers struggle with the most,
0:37:32 because there is no standard blueprint for it.
0:37:43 It’s a combination of an engineering problem, plus a data science problem, plus a regulation problem, plus winning consumer trust problem, because eventually you have to convince yourself and the stakeholders that this is safe enough.
0:37:49 So it’s not a science problem, but in many ways it’s still the blueprint is yet to be clearly written for this.
0:37:53 And I just don’t think the regulations and standards are enough and the same applies for cybersecurity.
0:37:57 I think they’re very short of what actually needs to be put into these systems.
0:38:01 What best practices do you use to reduce stress in your technical debt?
0:38:06 If you ask any seronix software engineer what the three magic words are, they’ll tell you this.
0:38:12 The kind of design principles that we follow is simple, correct, fast in that order.
0:38:19 The idea is that first of all, you want to build a very, very simple system that you understand from the electron level.
0:38:24 We do not integrate any technology that we don’t understand from the electron level all the way up to the software level.
0:38:32 And so the TLDR is that we build a ton of stuff that doesn’t work at first, but that we understand super, super well.
0:38:38 And so the systems design like kind of philosophy pervades basically everything in our software spec.
0:38:43 So yeah, the simple, correct, fast is like the way we do it at Seronic. This is totally worked from my opinion.
0:38:46 I never heard of technical debt, what’s that?
0:38:54 So I think technical debt is inevitable in fast changing industries like the ones we play in.
0:38:59 So I think we were previously talking about how the underlying technology in these eventual systems is changing,
0:39:04 which means we have to rethink the products we are providing to the market quite a bit, right?
0:39:09 So you keep the current system stable and then you deploy a new system magically over time.
0:39:14 But we’ve all been doing engineering for a long time to know that it’s never as clear cut as that.
0:39:20 But I think we focus a lot on working with our customers on like joint roadmaps where we literally go to customers
0:39:25 and say, hey, we’re going to actually upgrade the architecture of this product because now we can actually make use of generative AI
0:39:28 or any other technology in order to improve these products.
0:39:35 And surprisingly, more often than not, if you have a reasonable roadmap, the customers are actually pretty reasonable.
0:39:41 It does require you to have a trusted relationship where it’s just not like here’s some money and you’re a vendor and a traditional software vendor.
0:39:48 So if you’re in commoditized markets where you’re just purchasing based on certain features and who’s the most cheapest to procure,
0:39:50 it’s hard to have that kind of relationship.
0:39:57 But for the industries where and where software we are providing them, either that’s the developer tools or the actual software that’s going into the vehicle.
0:40:00 This is cutting edge software and they know this is cutting edge software.
0:40:04 So it’s a very like collaborative relationship with the customer.
0:40:09 We’re often building new products with them and that allows us to have sort of these thoughtful roadmaps.
0:40:11 Man, I’ve been in tech 25 years.
0:40:13 There’s always technical debt.
0:40:15 The longer you’re around, the more it piles up.
0:40:17 I don’t know that I could add anything more.
0:40:24 The one thing that’s a little bit unique about Skydio is that we have these hardware platforms and that allow us to basically sort of reset.
0:40:29 Like the latest generation that we announced last year was three years and $80 million in the making.
0:40:31 That allowed us to basically reset a lot.
0:40:37 But we also have a pretty large cloud software component and there’s a hell of a lot of technical debt that’s been built up over the last seven, eight years.
0:40:40 I don’t have a magical answer on that one.
0:40:46 Some people think to get to self driving level four or five or a robot taxi status that we don’t need lighter.
0:40:48 What’s your take there?
0:40:50 All right, I can take that super easy question.
0:40:56 So I think generally one is somehow the industry has gotten into this.
0:40:59 Yes, LiDAR, no LiDAR, sort of binary camps.
0:41:09 But at the end of the day, a sensor is just one other piece of technology that you incorporate to achieve the ultimate goal of, in this case, level four robot axes.
0:41:15 So if we more had a path to deploying the vehicles that you see on the road without a LiDAR,
0:41:23 I guarantee you we would have taken that path because there’s a lot of engineering and science invention that went into building some of the LiDARs that we more has.
0:41:25 They really are like cutting edge LiDARs.
0:41:31 And so we must not building them for the sake of just, hey, it’s fun to work on these technical challenges.
0:41:42 There’s a portion of the safety case that goes back to the question of how do you certify that these are safe enough that relies on certain properties that only LiDARs have that today.
0:41:45 The other combination of other sensors cannot.
0:41:50 And even in cars in like China, for example, you see LiDARs in lower levels of autonomy.
0:41:54 So A, that allows them to sort of bring certain capability to market faster.
0:42:07 But also from a liability perspective, you have to really think from a manufacturer perspective, the vehicle does get into an accident and we will see accidents happen in the industry as sort of the technology is maturing.
0:42:21 Would you rather have gone the all the way and done whatever was possible from a safety and liability perspective to prevent that or or not is the question that OEMs have to face when they think about should I put a LiDAR in a passenger car or not.
0:42:36 So, ultimately, will we get to a point where systems without LiDARs are good enough to function in their respective design domain that respective design domain could be mining that respective design domain could be cars.
0:42:40 Some defense applications, etc. I think in the long term, we will get there.
0:42:56 But do I think that at the early stages of deployment of technology, taking the safest path is the right approach in order to build up trust in the technology in order to build up trust with residents of a city in order to build up trust with regulators.
0:42:59 I think that’s a pretty reasonable approach.
0:43:03 I’m super curious to hear your observations around the ecosystems you’re seeing emerging.
0:43:06 I’m sure that’s also really interesting from a venture perspective, right?
0:43:09 How do you see the next couple of years coming together there?
0:43:11 I’m happy to take a first crack at this.
0:43:13 So, I think it’s a really good question.
0:43:14 It’s something that we think about a lot.
0:43:32 Historically, I’d say the industry has mostly focused around with a few exceptions applied being a great example of an exception scaly eye being another one that both businesses really took off as a result of selling software into the sort of self driving early on at least into the self driving market.
0:43:46 Most autonomy companies for the last decade have been very tight coupling of hardware and software where all of the software was written almost to the firmware level for the specific hardware form factor.
0:44:01 And what we’re starting to see, I think part of it is like the last three years of developments in AI and part of it is the broad explosion of interest in autonomy across a lot of different use cases is more of a software tool chain emerging.
0:44:08 But there are a lot of unsolved challenges that we see pretty consistently across companies operating in autonomous spaces.
0:44:18 So even like basic assumptions about how software is built for the kind of traditional SaaS world of the 2010s just doesn’t work in the autonomy space.
0:44:28 You often don’t have Wi-Fi like you can’t expect that you can just ship releases like every five minutes to customers and they’ll be able to update.
0:44:31 How do you stream data from one place to another?
0:44:42 Like how do you decide what decision making happens on device on chip on the edge versus like in a cloud environment because it has to consider a swarm or a fleet of devices.
0:44:54 So those are all like really interesting technical challenges and problems that I think developers will start solving more at scale in the next several years as more and more autonomy use cases come up.
0:45:03 So I imagine that the kind of developer toolkit around building an autonomous system is going to be more decoupled from the hardware itself over the next several years.
0:45:08 There’s lots of opportunities to build new tech in those scenes coming up.
0:45:13 From our vantage point, autonomy is what democratizes access to the drones, but it’s not what people buy.
0:45:15 They don’t care about our autonomy.
0:45:17 They frankly don’t really care about our drones.
0:45:26 What they care about is the information that the drones give them and the easier way we can get that information, the faster way we can get that information.
0:45:41 Our next level of autonomy in the broader roadmap is going to be less about how do you fly and more about what mission are we trying to accomplish and how do you actually coordinate across multiple drones to be able to accomplish that mission in a safe way in a fast way.
0:45:45 And that becomes much more vertically specific and use case specific.
0:45:50 And that’s really to us where like the gold is and will create tremendous differentiation.
0:45:56 Our hope is that there is actually a very thriving drone industry in the United States to be able to combat the Chinese manufacturers.
0:45:59 It’s also extraordinarily capital intensive and that’s hard.
0:46:00 That’s a big barrier.
0:46:05 We’ve raised a lot of money and we’re the largest manufacturer in the United States, but we still have a long, long, long way to go.
0:46:18 I hope there’s an sort of proliferation of a number of players coming in and contributing because I think it’ll be overall best for consumers is in building applications for the vehicle.
0:46:23 So today, for the most part, the cars we buy in US, in Europe, etc.
0:46:30 They’re not delightful consumer products like when you brought your first iPhone and you had a magical moment of ours is great.
0:46:35 And now some of that has, I think, tapered off at iPhone 16 seems like the same as 15 and 14.
0:46:40 So there’s really not been a great consumer product in that sense for a while.
0:46:46 But when you go to China and you experience those, you get some of that wow and that moment of delight back.
0:46:50 And then you peel the layers and say, OK, why doesn’t this exist in the cars here?
0:46:53 The reality is that you peel back the layers behind your car today.
0:47:02 There’s 150 different suppliers that each provided a small ECU as mini computer and the OEM integrated that into sort of a functional experience.
0:47:05 But you almost have to redesign the car from the ground up.
0:47:07 And that’s what Tesla did because they could start fresh.
0:47:11 They could start with a software engineering talent and say, we don’t need to design with the way that we did.
0:47:16 And that’s a journey setting every single vehicle that moves is on today.
0:47:20 And that’s why Applied is sort of providing that operating system.
0:47:25 Because if we can provide the operating system and give you a nice SDK to build those consumer applications on top,
0:47:29 you can unleash your creativity and think of the car as your third space.
0:47:31 And what experiences would you want in that vehicle?
0:47:34 Is it theater mode? Is it something else, et cetera.
0:47:37 And so that’s where I hope the industry goes.
0:47:40 And I think overall, it’ll be just better for consumers.
0:47:46 All right, that is all for today.
0:47:49 If you did make it this far, first of all, thank you.
0:47:53 We put a lot of thought into each of these episodes, whether it’s guests, the calendar touchers,
0:47:57 the cycles with our amazing editor, Tommy, until the music is just right.
0:48:03 So if you like what we’ve put together, consider dropping us a line at ratethespodcast.com/a16z.
0:48:05 And let us know what your favorite episode is.
0:48:08 It’ll make my day, and I’m sure Tommy’s too.
0:48:10 We’ll catch you on the flip side.
0:48:13 [music]
0:48:15 [music fades out]
0:48:17 you

0:00:02 This is the big race in robotics.
0:00:04 The smarter your brain, so to speak,
0:00:07 the less specialized your appendages have to be.
0:00:11 AI has pushed every single one of these kind of to its limit
0:00:13 and to a new state of the art.
0:00:14 The way they’re solving precision
0:00:17 is instead of throwing more sensors on the car,
0:00:19 is to basically throw more data at the problem.
0:00:22 Data is absolutely eating the world.
0:00:23 What is good enough?
0:00:24 We used to have the Turing test,
0:00:27 which obviously we’re blown past now.
0:00:30 His short hand for it was like the AWS of AI.
0:00:33 He’s got this idea of this distributed swarm
0:00:36 of unutilized inference computers.
0:00:39 Whether that’s an oil rig, whether that’s a mine,
0:00:41 whether that’s a battlefield,
0:00:43 there’s so many different use cases
0:00:46 for a lot of this underlying technology
0:00:49 that are really starting to see the light of day.
0:00:51 It’s basically an if not a win.
0:00:53 In inevitability.
0:00:54 Early this month,
0:00:58 Elon Musk and the team at Tesla held their We Robot event,
0:00:59 where they unveiled their plans
0:01:02 for the unsupervised full self-driving cyber cab
0:01:05 and RoboVan, plus Optimus,
0:01:08 their answer to consumer-grade humanoid robots,
0:01:10 and also what Musk himself predicted
0:01:14 would be, quote, “the biggest product ever of any kind.”
0:01:16 Now, of course, none of these products are on the market yet,
0:01:19 but several demos were on show at the event.
0:01:21 Naturally, the response was mixed.
0:01:23 Supporters said we got a glimpse of the future,
0:01:26 while critics said the details were missing.
0:01:29 But in today’s episode, we’re not here to debate that.
0:01:31 What we do want to talk about is what this indicates
0:01:35 about the intersection of where hardware and software meet.
0:01:38 So what does Rich Sutton’s 2019 blog post, The Bitter Lesson,
0:01:41 tell us about the decisions that Tesla’s making in autonomy?
0:01:45 And how realistic is the quoted $30,000 price range?
0:01:48 Also, what are the different layers of the autonomy stack?
0:01:50 And where do we get the data to power it?
0:01:51 And what does any of this look like
0:01:53 when you exit the consumer sphere?
0:01:56 We cover all this and more with A16C partners,
0:01:58 Anjani Midda and Aaron Price-Wright.
0:02:01 Anjani previously founded Ubiquiti 6,
0:02:04 a pioneering computer vision and multiplayer technology company
0:02:07 that sat right at this intersection of hardware and software,
0:02:10 and was eventually acquired by Discord.
0:02:13 Aaron, on the other hand, invests on our American dynamism team
0:02:15 with a focus on AI for the physical world.
0:02:17 And if you’d like to dig even deeper here,
0:02:20 Aaron has penned several articles on the topic
0:02:22 that we’ve linked in our show notes.
0:02:24 All right, let’s get to it.
0:02:29 As a reminder, the content here is for informational purposes only,
0:02:33 should not be taken as legal, business, tax, or investment advice,
0:02:35 or be used to evaluate any investment or security
0:02:38 and is not directed at any investors or potential investors
0:02:39 in any A16C fund.
0:02:42 Please note that A16C and its affiliates
0:02:46 may also maintain investments in the companies discussed in this podcast.
0:02:48 For more details, including a link to our investments,
0:02:51 please see a16c.com/dispotures.
0:03:00 So last week, Tesla had their WeRobot event
0:03:03 and Musk announced the CyberCab, the RoboVan,
0:03:07 or as he liked to call it, the RoboVan, and the Optimus.
0:03:10 You guys are so immersed in this hardware software world,
0:03:13 I’d love to just get your initial reaction.
0:03:16 From my perspective, it wasn’t that there was anything in particular
0:03:17 that was super surprising.
0:03:20 But what was exciting as just sort of a culmination
0:03:22 of one thing that Elon Musk does really well
0:03:24 and Tesla has done really well,
0:03:28 which is continue to pour love and energy and money and time
0:03:33 into a dream and a vision that’s been going on for a really long time,
0:03:36 like well past when most financial investors
0:03:39 and most people kind of lost the luster of self-driving cars
0:03:44 after their initial craze in the mid to late 2010s.
0:03:46 And they’ve just continued to plot along
0:03:48 and to continue to make developments
0:03:51 and now we’re finally seeing this glimpse of the future
0:03:53 for the first time in a really long time.
0:03:57 I think that’s right. I think it was very impressive, but unsurprising.
0:03:58 Yeah.
0:04:01 So I think the two schools of thought when people watched the event was,
0:04:05 one was absolutely this whole, “Oh my God, this is such vaporware.”
0:04:08 He shared literally nothing on engineering details.
0:04:12 “What the hell? Come on, give us the meat on timelines and dates and prices.”
0:04:16 And then the opposing view was like, “Holy shit. They’re still going.
0:04:18 They haven’t given up on any of this autonomy stuff
0:04:19 that he’s been talking about for years.”
0:04:22 And I’m absolutely more empathetic towards the view that it was a lean towards the ladder,
0:04:25 which is that it saw it as an homage to the bitter lesson.
0:04:27 It’s sort of amazing blog post that I’m going to do a terrible job
0:04:30 of summarizing by this great computer scientist, Rick Sutton,
0:04:34 which basically says that over the last 70 years or so of computer science history,
0:04:36 what we’ve learned is that general purpose methods
0:04:40 basically beat out any specific methods in artificial intelligence,
0:04:43 in particular, basically the idea that if you’re working on solving a task
0:04:47 that requires intelligence, you’re usually better off leveraging Moore’s law
0:04:52 and more compute and more data than trying to hand engineer a technique
0:04:54 or a set of algorithms to solve a particular task.
0:04:57 And broadly speaking, that’s been the big grand debate
0:04:59 in self-driving and autonomy I would say for the last two decades, right?
0:05:03 This is the sort of general purpose bitter lesson school versus the,
0:05:06 let’s model self-driving as a specific task.
0:05:11 As a set of discreet decision-making algorithms unconnected to each other.
0:05:15 A system to solve, let’s say, edge detection around stop signs, right?
0:05:18 Whereas self-driving is a really hard problem.
0:05:20 And you could totally say, well, there’s so many edge cases in the world
0:05:23 that we should map out each of those edge cases.
0:05:25 And I think it was an homage to the bitter lesson.
0:05:28 So that’s what I was most excited about is he did share actually details
0:05:31 that their pipeline is basically an end-to-end deep learning approach.
0:05:34 Which is incredible and probably true only for the last,
0:05:37 my guess is 18 to 24 months, right?
0:05:39 Yeah. And I mean, in the bitter lesson,
0:05:44 he also talks about the fact that it’s really appealing to do the opposite.
0:05:46 Because in the short term, you will get the benefit,
0:05:51 but the broader deep learning approach ends up winning out in the long term.
0:05:54 And a lot of people talk about Musk Musk says about himself
0:05:56 that the timelines sometimes are off,
0:05:59 but he’s basically banking on that premise in the long term.
0:06:02 It’s basically an if not a win.
0:06:03 In inevitability.
0:06:09 And I think the event was the first time that it really did feel,
0:06:13 in an emotional sense, for the average American consumer.
0:06:16 I’m not talking about the super-duper tech literate people
0:06:18 who wanted the details of the underlying models and their weights,
0:06:21 but like for the average American consumer,
0:06:25 the first time that this version of the future felt like in inevitability.
0:06:29 And before we get into maybe the specifics around where else hardware
0:06:31 and software are intersecting,
0:06:34 I’d love to just talk about that, that average person who’s watching,
0:06:37 because you guys are meeting with companies and investors,
0:06:40 and this has been going on for quite some time.
0:06:43 So I’m just curious if maybe you noticed anything under the hood,
0:06:47 or maybe the meta in that announcement or event,
0:06:50 that maybe the average person watching is, you know, what are they seeing?
0:06:53 They’re saying things like, oh, maybe it was human-controlled
0:06:55 and not like fully AI and device,
0:06:58 or other people are commenting on the fact that these humano
0:07:01 are shaped like humans, like why do we need that?
0:07:03 On the topic of humanoids,
0:07:08 I think humanoids are a great choice of embodiment for a robot
0:07:13 to really emotionally connect to and speak to a human being watching,
0:07:16 because I can relate to a human form factor.
0:07:18 Obviously, we found out that it was teleoperative,
0:07:23 which is in my opinion still doesn’t take away from like how cool and amazing it was.
0:07:28 The human form factor is a way to connect what is happening with robotics
0:07:31 to a regular person who is like, okay, yes, I like see myself in that.
0:07:34 This looks like Star Wars or some other sci-fi movie.
0:07:37 In reality, maybe this is like a controversial opinion.
0:07:41 I don’t see the vast majority of economic impact over the next decade
0:07:44 from robotics coming from the humanoid form factor,
0:07:47 but that doesn’t take away from the power of the symbol
0:07:50 of having a humanoid make a drink at this event,
0:07:55 because it just like connects back to this sort of science fiction promise
0:07:59 of our childhoods getting sort of finally delivered.
0:08:02 The opening sequence, he started with like a sci-fi,
0:08:04 I think he was a Blade Runner visual and he was like,
0:08:06 we all love sci-fi and I want to be wearing that jacket
0:08:07 that he’s wearing in the picture,
0:08:09 but we don’t want any of the other dystopian stuff.
0:08:13 And so that definitely stuck out to me is that he did not start
0:08:14 the way he usually does.
0:08:16 It’s often a technical first sort of story,
0:08:20 but he started with here’s a vision for where I think the world should go.
0:08:22 So it was much more Disney-esque in that and it was quite poetic.
0:08:25 I think they literally did it on the Warner Brothers Studio lot.
0:08:28 And so they like recreated a bunch of cities
0:08:31 and I think they had on site at the event the robovans
0:08:34 taking people around from these simulated cities.
0:08:37 There was a sort of theatricality to it all that stuck out to me,
0:08:38 which I thought was quite different.
0:08:40 And I thought it was refreshing
0:08:43 because the core problem with this branch of AI,
0:08:46 which is largely deep learning based and bitter lesson based,
0:08:47 is that it’s an empirical field.
0:08:50 Unlike, call it Moore’s Law, which was predictive,
0:08:53 where you basically know if you double that number of transistors,
0:08:55 you get this much more performance on the chip.
0:08:56 And it’s just about pure execution.
0:08:58 AI is much more empirical.
0:09:01 You don’t really know when the model is going to get done training
0:09:03 and when it does get training, whether it will converge or not.
0:09:06 Or even what is converging mean, like what is good enough?
0:09:10 We used to have the Turing test, which obviously we’re blown past now.
0:09:13 It’s a feeling more than it is a set of discrete metrics
0:09:14 that you can really point to.
0:09:15 Right.
0:09:19 So it made a lot of sense to me that he’s trying to decouple
0:09:23 this idea of progress from a specific timeline.
0:09:23 I see.
0:09:25 Because I just think we’re setting ourselves up
0:09:28 for every time you ask a deep learning researcher.
0:09:30 So when’s that GPT-5 model going to show up?
0:09:32 It’s like the most frustrating question ever, right?
0:09:34 Because they don’t know.
0:09:34 We don’t know.
0:09:37 And frankly, sometimes they show up earlier than schedule
0:09:38 and show up later.
0:09:40 And by the way, you can look at the stock market’s reaction.
0:09:43 It’s a prime example of how people have been so conditioned by,
0:09:47 I would say, the Steve Jobsian, Apple-like cadence year on year of like,
0:09:50 here’s your new iPhone, it’s incremental but predictable.
0:09:53 I think forecasting that the tech industry keeps trying to reward.
0:09:55 And I think what he’s doing is pretty refreshing, which is saying,
0:09:58 look, here’s a vision for where we want to go.
0:09:59 But it’s decoupled.
0:10:01 The second thing on the humanoid piece that I was quite impressed by
0:10:04 is actually the quality of the tele-operation.
0:10:06 So everybody’s talking about how, oh, this is fake.
0:10:07 This is all smoke and mirrors.
0:10:08 It’s just people.
0:10:10 Tele-operation is real.
0:10:12 I was going to say, why is no one talking about that?
0:10:13 Have you ever tried?
0:10:14 I mean, I’ve tried.
0:10:15 It’s so hard.
0:10:19 We were at a company two weeks ago and they’ve got these tele-op robots.
0:10:25 And the founder was demoing a mechanical arm that he was tele-operating with a gamepad.
0:10:27 And he was folding clothes with it.
0:10:28 And I was like, oh, that looks simple.
0:10:30 He’s like, here, try it.
0:10:32 It was one of the hardest manipulation things I’ve ever tried.
0:10:37 And by the way, we tried that with VR headset, with six-stop motion controllers,
0:10:39 almost harder to do.
0:10:42 Tele-operating something, especially over the internet,
0:10:45 in a smooth fashion with precision is incredibly hard.
0:10:49 And I don’t think people appreciate the degree to which they’ve really solved that pipeline.
0:10:50 Yeah, I was actually really impressed by that.
0:10:56 And I think that there’s huge opportunity for tele-op in production applications
0:10:59 that will have massive economic benefit,
0:11:03 even before we have true robots running around managing themselves.
0:11:07 Because if you think about there’s all these really hard and really dangerous
0:11:10 or hard to get to jobs, or there’s labor differentials
0:11:14 where it’s a lot harder to hire people to do certain things in certain locations.
0:11:19 And if we can imagine a future where the tele-op that we saw last week at the event
0:11:22 is something that’s widely available, that’s incredible.
0:11:25 Imagine not having to go and service a power line,
0:11:29 but you can actually tele-op a robot to do that for you,
0:11:31 but still have the level of human training and precision needed
0:11:35 to make a really detailed and specific evaluation.
0:11:38 The promise of that is really cool, even before we get to robots.
0:11:40 So that was really exciting.
0:11:42 Yeah, it’s like a stop along this journey.
0:11:46 And so if we talk about that journey, the arc of hardware and software coming together
0:11:49 in maybe a different way than we’ve seen in the past.
0:11:52 Just as an example, so Mark famously said, software is eating the world.
0:11:54 That was in 2011 or in 2024.
0:11:58 And it does feel like the last decade has been a lot of traditional software,
0:12:02 not so much integrating with the physical world around us.
0:12:05 And so where would you place us in that trajectory?
0:12:07 Because we’re seeing it with autonomous vehicles,
0:12:10 but I get the sense that’s not the only place where this is happening.
0:12:15 Yeah, this is where I spend 95% of my time in all of these industries
0:12:19 that are just starting to see the glimmers of what autonomy
0:12:22 and sort of software-driven hardware can bring.
0:12:25 What’s really interesting is just actually a dearth of skills
0:12:29 of people who know how to deal with hardware and software together.
0:12:31 You have a lot of people that went and got computer science degrees
0:12:34 over the last decade, and relatively speaking,
0:12:38 a lot fewer than when got electrical engineering or mechanical engineering degrees.
0:12:39 And we’re starting to see the rise of, oh, shoot,
0:12:43 we actually need people who understand not just maybe how the software works.
0:12:49 In the cloud with Wi-Fi, where you have unlimited access to compute,
0:12:51 and you can retry things as many times as you want,
0:12:54 and you can ship code releases all day every day.
0:12:57 But you actually have kind of a hardware deployment,
0:12:59 where you have limited compute in an environment
0:13:02 where you maybe can’t rely on Wi-Fi all the time,
0:13:04 where you have to tie your software timelines
0:13:06 to your hardware production timelines.
0:13:10 Like, these are a really difficult set of challenges to solve.
0:13:13 And right now, there just isn’t a lot of standardized tooling
0:13:15 for developers and how to do that.
0:13:19 So it’s interesting, we’re starting to see portfolio companies of ours
0:13:22 across really different industries that are trying to use autonomy,
0:13:27 whether it’s oil and gas or water treatment or HVAC or defense.
0:13:31 They’re like sharing random libraries that they wrote
0:13:33 to connect to, like, particular sensor types.
0:13:36 Because there’s not this, like, rich ecosystem of tooling
0:13:38 that exists for the software world.
0:13:42 So we’re really excited about what we’re starting to see emerge in the space.
0:13:44 Even Elon said when he’s talking about these two different products
0:13:47 that he’s unveiling, right, Optimus,
0:13:50 and then you have the RoboVans or CyberCabs.
0:13:53 And those seem like two completely different things,
0:13:55 but he even said in the announcement,
0:13:56 he said, “Everything we’ve developed for our cars,
0:13:59 the batteries, power electronics, advanced motors,
0:14:03 gearboxes, AI inference computer, it all applies to both.”
0:14:04 Right, so you’re seeing this overlap.
0:14:05 That’s super exciting.
0:14:06 When I was watching it, I was just nerding out
0:14:09 because my last company was a computer vision 3D mapping
0:14:10 and localization company.
0:14:14 So I unfortunately spent too much of my life calibrating LiDAR sensors
0:14:15 to our computer vision sensors.
0:14:18 Because our whole thesis when I started back in 2017
0:14:21 was that you could do really precise positioning
0:14:22 just off of computer vision.
0:14:25 And that you didn’t need fancy hardware like LiDARs or depth sensors.
0:14:27 And to be honest, not a lot of people thought that we could pull it off.
0:14:30 And frankly, I think there were moments when I doubted that too.
0:14:33 And so it was just really fantastic to see that his bet
0:14:36 and the company’s bet on computer visions
0:14:38 and a bunch of these sensor fusion techniques
0:14:41 that would not need specialized hardware
0:14:43 would ultimately be able to solve
0:14:45 a lot of the hard navigation problems,
0:14:48 which basically means that the way they’re solving precision
0:14:51 is instead of throwing more sensors on the car,
0:14:53 is to basically throw more data at the problem.
0:14:56 And so in that sense, data is absolutely eating the world.
0:14:59 And you asked, where on the trajectory are we of soft-rating the world?
0:15:01 And I think we’re definitely on an exponential
0:15:03 that has felt like a series of stacked sigmoids.
0:15:05 Often it feels like you’re on a plateau.
0:15:08 But a series of plateaus totally make up an exponential
0:15:09 if you zoom out enough.
0:15:11 And earlier in the conversation we talked about the bitter lesson,
0:15:13 a number of other teams in the autonomy space
0:15:15 decided to tackle it as a hardware problem,
0:15:16 not a software problem, right?
0:15:17 Where they said, well…
0:15:22 More LiDAR, more expensive LiDAR, more GPUs, more GPUs.
0:15:25 And Elon’s like, you know, actually I want cheap cars
0:15:27 that just have computer vision sensors.
0:15:30 And what I’m going to do is use a bunch of the custom,
0:15:34 really expensive sensors that many other companies put on the car,
0:15:35 which is at inference time.
0:15:37 And he’s just going to use them at train time.
0:15:40 So Tesla does have a bunch of like really custom hardware
0:15:43 that’s not scalable, that drives around the world
0:15:45 in their parking lots and simulation environments and so on.
0:15:49 And then they distill the models they train on that custom hardware
0:15:50 to a test time package.
0:15:53 And then they send that test time package to their retail cars,
0:15:55 which just have computer vision sensors.
0:15:57 And the reality is that’s a raw arbitrage, right?
0:15:58 Between sensor stack.
0:16:02 And it allows the hardware out in the world to be super cheap.
0:16:04 The result there is software is eating the sensor stack
0:16:08 out in the world that makes the cost of these cars so much cheaper
0:16:11 that you can have a $30,000 fully autonomous car
0:16:14 versus $100 plus thousand dollars of cars
0:16:17 that are fully loaded with these LiDAR sensors and so on.
0:16:19 But I think in order to have the intuition
0:16:20 that you can even do that,
0:16:23 you really actually have to understand hardware.
0:16:26 If you just understand software and hardware
0:16:30 is like a sort of a scary monster that lives over here
0:16:32 and maybe you have a special hardware team that does it,
0:16:35 it’s going to be hard for you to have the confidence to say,
0:16:37 “No, we can do it this way.”
0:16:38 I think you’re totally right,
0:16:41 which is that the superpower that Tesla has
0:16:44 is his ability to go full stack, right?
0:16:46 Because a lot of other industries
0:16:49 often segment out software versus hardware like you’re saying.
0:16:51 And that means that people working on algorithms
0:16:55 and the autonomy part just treat hardware as like an abstraction, right?
0:16:57 You throw over a spec, it’s an API,
0:16:59 it’s an interface that I program against
0:17:00 and I have no idea what’s going on.
0:17:02 You don’t have to worry about the details, it doesn’t matter.
0:17:03 Which by the way is super powerful.
0:17:06 It’s unlocked this whole general purpose wave
0:17:09 of models like ChatGPT and so on, right?
0:17:11 Because it allows people who specialize in software
0:17:13 to not have to think about the hardware.
0:17:16 It’s also what’s driven sort of the software renaissance
0:17:17 of the last 15 years.
0:17:18 Absolutely, decoupling, right?
0:17:20 Composition and abstraction is sort of the fundamental basis
0:17:22 of the entire computing revolution.
0:17:24 But I think when you’re like him
0:17:26 and you’re trying to bring a new device to market,
0:17:28 kind of like what Jobs did with the iPhone,
0:17:29 by going full stack,
0:17:32 you end up unlocking massive efficiencies of cost.
0:17:35 And I think that’s what this event
0:17:38 may have been lost in the sort of theatricality of it all
0:17:40 is the fact that he’s able to deliver
0:17:43 an autonomous device to retail consumers
0:17:45 at a cost profile through vertical integration
0:17:47 that would just not be possible
0:17:48 if it was just a software team
0:17:50 buying hardware from somebody else and building on top.
0:17:52 Can we talk about those economics, by the way?
0:17:54 Just attacking that head on.
0:17:56 Both Optimus and CyberCAD were quoted
0:17:59 as being under the 30K range.
0:18:00 Is that really realistic?
0:18:02 And then tied into what you were saying,
0:18:04 we see other autonomous vehicles
0:18:07 which are betting more on the lidar and the sensors,
0:18:11 which also have come down in price pretty substantially.
0:18:14 My guess is Elon is backing into the cost
0:18:16 based on what people are willing to pay.
0:18:18 And he will do whatever it takes
0:18:19 to get those costs to line up.
0:18:21 I mean, it’s the same thing he did with SpaceX.
0:18:24 He will operate within whatever cost constraints
0:18:25 he needs to operate within,
0:18:26 even if the rest of the market
0:18:29 or the research community is telling him
0:18:30 it’s not possible.
0:18:33 Obviously, like a 30K humanoid robot
0:18:37 is way less than what most production
0:18:39 industrial robotic arms cost today,
0:18:41 which I think are more in the 100K range
0:18:43 for the ones that are used in like the high end factory.
0:18:47 So if he can get it down to 30K, that’s really exciting.
0:18:48 I also don’t necessarily think
0:18:50 you need even a 30K humanoid robot
0:18:53 to accomplish a wide swath of the automation tasks
0:18:55 that would pretty radically transform
0:18:57 the way our economy functions today.
0:18:58 Yeah, I think Aaron’s right
0:19:00 in that there’s probably a top-down director
0:19:01 who just do whatever it takes
0:19:02 to get into the cost footprint of–
0:19:04 This car has to cost 30K.
0:19:04 Right.
0:19:06 But I think if you do a bottoms-up analysis,
0:19:07 I don’t think you end up too far
0:19:09 because actually if you just break down
0:19:11 the kind of bomb on a Tesla Model 3,
0:19:13 you’re not dramatically far off
0:19:14 from the sensor stack you need
0:19:17 to get to a $30,000 car, right?
0:19:19 This is the beauty of solving
0:19:20 your hardware problems with software
0:19:24 is you don’t need a $4,000 scanning lighter
0:19:25 on the car.
0:19:28 So I think on the CyberCab,
0:19:29 I feel much more confident
0:19:32 that the cost footprint is going to fall in that range
0:19:33 because it’s frankly–
0:19:35 We kind of have at least an ancestor on the streets, right?
0:19:38 The thing that gets prices up is custom sensors
0:19:40 because it’s really expensive
0:19:43 to build custom sensors in short production runs.
0:19:45 And so you either have scale of manufacturing
0:19:47 like an Apple and you make a new CMOS sensor
0:19:49 or a new Face ID sensor
0:19:51 and you get cost economies of scale
0:19:53 because you’re shipping more like 30 million devices
0:19:54 in your first run.
0:19:56 Or you just lean on commodity sensors
0:19:57 from the past era
0:19:59 and then you tackle most of your problems in software,
0:20:01 which is what he’s doing.
0:20:03 And to that point, when he’s betting on software,
0:20:05 another interesting thing that he announced
0:20:07 was really over-specking these cars
0:20:09 to almost change the economics potentially
0:20:11 based on the fact that those cars
0:20:13 could be used for distributed computing.
0:20:13 To your point, Ange,
0:20:17 if you put a bunch of really expensive sensors on the car,
0:20:20 you can’t really distribute the load of that
0:20:23 in any other way than driving the car, right?
0:20:25 But if you actually have this computing layer
0:20:27 that’s again, in his case,
0:20:29 he’s saying he’s planning to over-spec,
0:20:30 that actually can fundamentally change
0:20:32 like what this asset is.
0:20:34 And you kind of saw the same thing
0:20:35 even with Tesla’s today
0:20:37 where he’s talking about this distributed grid, right?
0:20:39 Where all of a sudden these large batteries
0:20:41 are being used not just for the individual asset.
0:20:43 So do you have any thoughts on that idea
0:20:44 or if we’ve seen that elsewhere?
0:20:46 He was a bit skimpy on details on that.
0:20:47 Of course.
0:20:50 But I think he did say that the AI5 chip is over-spec’d.
0:20:54 It’s probably going to be four to five times more powerful
0:20:56 than the HW4, which is their current chip.
0:20:58 It’s going to draw four times more power,
0:21:01 which probably puts in that like 800 watts or so range,
0:21:02 which for context,
0:21:05 your average hair dryer is at about 1800 watts.
0:21:07 I mean, it’s hard to run power on the edge.
0:21:11 But I think what he said was something to the effect of like,
0:21:14 your car’s not working 24 hours a day.
0:21:17 So if you’re driving, call it eight hours a day in LA traffic.
0:21:19 God bless whoever’s having to do that.
0:21:20 For real.
0:21:21 Hopefully they’re using self-driving.
0:21:22 One would hope.
0:21:24 Actually, he opened up his pitch with a story
0:21:25 about driving to El Segundo
0:21:27 and he’s saying you can fall asleep
0:21:28 and wake up on the other side.
0:21:30 But I think the t-shirt size he gave
0:21:37 was about 100 gigawatts of unallocated inference compute
0:21:38 just sitting out there in the wild.
0:21:41 And I think his shorthand for it was like the AWS,
0:21:42 of AI, right?
0:21:45 He has got this idea of this distributed swarm
0:21:48 of unutilized inference computers.
0:21:50 And it’s a very sexy vision.
0:21:51 I really want to believe in it.
0:21:53 Ground us on, is this realistic?
0:21:55 Well, I think it’s realistic for workloads
0:21:58 that we don’t know yet in the following sense, right?
0:22:00 That the magic of AWS is that it’s centralized
0:22:03 and it abstracts away a lot of the complexity
0:22:05 of hardware footprints for developers.
0:22:08 And by centralizing almost all their data centers
0:22:09 in single locations with very cheap power
0:22:12 and electricity and cooling,
0:22:14 these clouds are able to pass on very cheap inference costs
0:22:15 to the developer.
0:22:16 Now, what he’s got to figure out is
0:22:19 how do you compensate for that in a decentralized fashion?
0:22:20 And I think we have kind of prototypes of this today.
0:22:23 Like, they’re these vast decentralized clouds.
0:22:24 I think one is literally called vast
0:22:26 of people’s unallocated gaming rigs.
0:22:31 People have millions of Nvidia 4090 gaming cards
0:22:32 sitting on their desks that aren’t used.
0:22:34 And historically, those have not yet turned
0:22:37 into great businesses or high-utilized networks
0:22:39 because developers are super sensitive
0:22:42 to do things, cost and reliability.
0:22:44 And by centralizing things,
0:22:47 AWS is able to ensure very high uptime and reliability,
0:22:50 whereas somebody’s GPU sitting on their…
0:22:52 Maybe available, maybe they’re driving to Elson now.
0:22:53 Right, they’re right.
0:22:54 And there are just certain things,
0:22:56 especially with AI models that are hard to do
0:22:58 on highly distributed compute
0:23:00 where you actually need good interconnect
0:23:03 and you need things to be reasonably close to each other.
0:23:03 Maybe in his vision,
0:23:06 there’s a world where you have optimist robots
0:23:10 in every home and somehow your home optimist robot
0:23:12 can take advantage of additional compute
0:23:14 or additional inference with your Tesla car
0:23:16 that’s sitting outside in your driveway.
0:23:17 Who knows?
0:23:18 Right.
0:23:20 Okay, well, this event clearly was focused
0:23:24 on different models that are consumer-facing.
0:23:25 So again, CyberCab,
0:23:27 that’s for someone using an autonomous vehicle.
0:23:30 Optimist is a human-eyed robot probably in your home.
0:23:33 But, Erin, you’ve actually been looking
0:23:36 at the hardware software intersection
0:23:37 in a bunch of other spaces, right?
0:23:39 And as you alluded to earlier,
0:23:41 maybe different applications
0:23:43 with better economics at least today.
0:23:45 I think long-term,
0:23:48 there’s no market bigger than the consumer market.
0:23:51 So everyone having a robot in their home
0:23:54 and a Tesla car in their driveway
0:23:57 that’s also a robot taxi has huge economic value.
0:23:59 But that’s also a really long-term vision
0:24:02 and there’s just so much happening in autonomy
0:24:05 that’s taking advantage of the momentum
0:24:07 and the developments that companies like Tesla
0:24:09 have put forward into the world over the last decade
0:24:11 that actually have the potential
0:24:15 to have meaningful impact on our economy in the short term.
0:24:17 I think the biggest broad categories for me
0:24:21 are largely the sort of dirty and unsexy industries
0:24:24 that have very high cost of human labor,
0:24:28 often because of safety or location access,
0:24:32 whether that’s an oil rig out in the middle of Oklahoma
0:24:35 somewhere that’s three hours drive from New York City,
0:24:38 whether that’s a mine somewhere in rural Wyoming
0:24:40 that freezes over for six months out of the year
0:24:43 so humans can’t live there and mine,
0:24:46 whether that’s a battlefield where you’re starting to see
0:24:49 autonomous vehicles go out and clear bombs and mines
0:24:51 from battlefields to protect human life.
0:24:54 There’s so many different use cases
0:24:56 for a lot of this underlying technology
0:24:58 that are really starting to see the light of day.
0:25:00 So very excited about that.
0:25:02 And as we think about that opportunity,
0:25:05 you’ve also talked about this software-driven autonomy stack.
0:25:07 So as you think about the stack, what are the layers?
0:25:09 Can you just break that down?
0:25:09 Yeah, sure.
0:25:12 So across whether it’s a self-driving car
0:25:14 or sort of an autonomous control system,
0:25:19 we’re seeing the stack break down into pretty similar categories.
0:25:20 So first is perception.
0:25:23 You have to see the world around you,
0:25:25 know what’s going on, be able to see if there’s a trash can,
0:25:28 be able to understand if there’s a horizon, if you’re a boat.
0:25:30 The second is something Anjno’s really well,
0:25:32 which is location and mapping.
0:25:33 So, okay, what do I see?
0:25:36 How do I find out where I am within that world
0:25:39 based on what I can see and what other sensors I can detect,
0:25:42 whether it’s GPS, which often isn’t available
0:25:46 in battlefields or in warehouses, et cetera.
0:25:47 The third is planning and coordination.
0:25:52 So that’s, okay, how do I take a large task
0:25:55 and turn it into a series of smaller tasks?
0:25:58 So what is more of an instant reaction?
0:26:02 I don’t have to really think about how to take a drink of water,
0:26:07 but I might have to think about how to make a glass of lemonade from scratch.
0:26:12 So how do I think about compute across those different types of regimes
0:26:13 when something is more of an instinct,
0:26:16 versus when something has to be sort of taken down
0:26:19 and processed into discrete operations?
0:26:20 And then the last one is control.
0:26:23 So that’s like, how does my brain talk to my hand?
0:26:26 Like, how do I know what are the nerve endings doing
0:26:29 in order to pick up this water bottle and take a drink out of it?
0:26:31 And that’s a really interesting kind of field
0:26:34 that’s existed for decades and decades,
0:26:36 but for the first time, probably since the ’70s,
0:26:38 we’re starting to see really interesting stuff happen
0:26:41 in the space of controls around autonomy and robotics.
0:26:45 And I would say like, all of these are pre-existing areas.
0:26:49 None of this is wildly new, but I think in the last two years,
0:26:51 especially with everything that’s happening
0:26:53 with deep learning video language models,
0:26:57 broadly speaking, AI has pushed every single one of these
0:27:00 kind of to its limit and to a new state of the art.
0:27:03 And there just aren’t tools that exist to tie all that together.
0:27:05 So every single robotics company,
0:27:06 every single autonomous vehicle company
0:27:10 is basically like rebuilding this entire stack from scratch,
0:27:13 which we see as investors as a really interesting opportunity
0:27:15 as the ecosystem evolves.
0:27:17 And as you think about that ecosystem,
0:27:19 people kind of say that as soon as you touch hardware,
0:27:21 you’re literally working on hard mode
0:27:23 compared to just a software-based business.
0:27:25 So what are the unique challenges?
0:27:29 Even with maybe that AI wave today that’s pushing things ahead,
0:27:32 how would you break down what becomes so much harder?
0:27:34 I think Ange touched on this a little bit before,
0:27:37 but the more you can commoditize the hardware stack, the better.
0:27:39 So the most successful hardware companies
0:27:42 are the ones that aren’t necessarily inventing a brand new sensor,
0:27:45 but are just taking stuff off the shelf and putting it together.
0:27:47 But still like tying everything together is really hard.
0:27:51 Like when you think about releasing a phone, for example,
0:27:53 Apple has a pretty fast shipping cadence,
0:27:57 and they’re still releasing a new phone only every once a year.
0:28:02 So you have to essentially tie a lot of your software timelines
0:28:05 to hardware timelines in a way that doesn’t exist
0:28:08 when you can just sort of ship whenever you want in a cloud.
0:28:09 If you need a new sensor type,
0:28:12 or you need a different kind of compute construct,
0:28:14 or you need something fundamentally different in the hardware,
0:28:16 you’re bound by those timelines.
0:28:19 You’re bound by your manufacturer’s availability.
0:28:24 You’re bound by how long it takes to quality engineer and test a product.
0:28:26 You’re bound by supply chains.
0:28:29 You’re bound by figuring out how these things have to integrate together.
0:28:34 So the cycles are often just quite a lot slower.
0:28:37 And then the other thing is when you’re interacting with the physical world,
0:28:41 you get into use cases that touch safety in a really different way
0:28:44 than we think about with pure software alone.
0:28:49 And so you have to design things for a level of hardiness and reliability
0:28:52 that you don’t always have to think about with software by itself.
0:28:55 If your chat GBT is a little slow, it’s fine.
0:28:56 You can just try again.
0:28:57 But if you have an autonomous vehicle
0:29:02 that’s like driving a tank on a battlefield autonomously,
0:29:05 and something doesn’t work, you’re kind of screwed.
0:29:09 So you have to have a much higher level of rigor and testing and safety
0:29:13 built into your products, which slows down the software cycles.
0:29:18 The Holy Grail is sort of general purpose intelligence for robotics,
0:29:22 which we still don’t have when you train a general model.
0:29:26 You basically get the ability to build hardware systems
0:29:28 that don’t have to be particularly customized.
0:29:31 And that reduces hardware iteration cycles dramatically.
0:29:33 Because you can basically say, look, roughly speaking,
0:29:36 these are the four or five commodity sensors you need.
0:29:38 The smarter your brain, so to speak,
0:29:41 the less specialized your appendages have to be.
0:29:43 And I think what a number of really talented teams
0:29:44 are trying to solve today is,
0:29:47 can you get models to generalize across embodiments?
0:29:50 Can you train a model that can work seamlessly
0:29:53 on a humanoid form factor or a mechanical arm,
0:29:55 a quadruped, whatever it might be?
0:29:57 And I’m quite bullish that it will happen.
0:29:59 I think the primary challenge there
0:30:00 that teams are struggling with today
0:30:03 is the lack of really high quality data.
0:30:06 The big unknown is just how much data,
0:30:08 both in quantity and quality,
0:30:10 do you really need to get models to be able to reason
0:30:13 about the physical world spatially
0:30:16 in a way that abstracts across any hardware?
0:30:18 I’m completely convinced that once we unlock that,
0:30:20 the applications are absolutely enormous.
0:30:23 Because it frees up hardware teams,
0:30:23 like Aaron was saying,
0:30:26 from having to couple their software cycles
0:30:27 from hardware cycles.
0:30:29 It decouples those two things.
0:30:30 And I think that’s the holy grail.
0:30:34 I think what Tesla, the victory of the autonomy team over there,
0:30:37 is having realized eight years ago
0:30:40 the efficacy of what we call early fusion foundation models,
0:30:42 which is the idea that you take a bunch of sensors
0:30:46 at training time and different inputs of vision, depth.
0:30:48 You take in video, audio.
0:30:51 You take a bunch of different six-doff sensors
0:30:52 and you tokenize all of those
0:30:55 and you fuse them all at the point of training.
0:30:57 And you build an internal representation
0:30:59 of the world for that model.
0:31:02 In contrast, the LLM world does what’s called late fusion.
0:31:04 You often start with a language model.
0:31:05 Let’s train just on language data
0:31:07 and then you duct tape on these other modalities,
0:31:09 like image and video and so on.
0:31:11 And I think the world has now started to realize
0:31:13 that early fusion is the way forward,
0:31:16 but of course they have an eight-year head start.
0:31:17 And so I get really excited when I see teams
0:31:20 either tackling the data sort of challenge
0:31:22 for general spatial reasoning,
0:31:24 or teams that are taking these early fusion
0:31:26 foundation model approaches to reasoning
0:31:29 that then allow the most talented hardware teams
0:31:31 to focus really on what they know best.
0:31:34 Where are these companies getting training data?
0:31:36 Because you mentioned Tesla, for example.
0:31:38 Yes, we’ve had cars on the road,
0:31:40 tons of them with these cameras and sensors.
0:31:44 I still think that one of the smartest things Elon did
0:31:47 was turn on full self-driving for everybody
0:31:50 for like a month-long trial period last summer.
0:31:53 I have a Tesla and I turned it on for my free month.
0:31:56 And it was like a life-changing experience using.
0:31:58 And I obviously couldn’t get rid of it.
0:32:00 And so now, not only do I now pay for full self-driving,
0:32:01 but I also–
0:32:02 You’re feeding the pipeline.
0:32:04 I’m giving him all my data.
0:32:06 So to me, that’s really clever.
0:32:08 And so I’m curious if you talk about
0:32:10 some of these other applications.
0:32:13 Do they have the number of devices,
0:32:16 or in this case, cars for Tesla capturing this data?
0:32:19 Or how else are we going to get this spatial data?
0:32:22 This is the big race in robotics right now.
0:32:24 I think there are several different approaches.
0:32:27 Some people are trying to use video data for training.
0:32:30 Some people are investing a lot in simulation
0:32:33 and creating digital 3D world.
0:32:37 And then there’s a mad rush for every kind of generated data
0:32:38 that you could possibly have.
0:32:41 So whether that’s robotic tele-operated data,
0:32:44 whether that’s robotic arms and offices,
0:32:48 most of these robotics companies have pretty big outposts
0:32:50 where they’re collecting data internally.
0:32:52 They’re giving humanoids to their friends
0:32:53 to have in their homes.
0:32:55 It’s a yes and scenario right now
0:32:58 where everyone is just trying to get their hands on data
0:32:59 literally however they can.
0:33:00 I think it’s the Wild West.
0:33:03 But if you’re Tesla, then the secret weapon you have
0:33:05 is you’ve got your own factories, right?
0:33:08 So the Optimus team has a bunch of humanoids
0:33:09 walking around the factories,
0:33:11 constantly learning about the factory environment.
0:33:13 And that gives them this incredible self-fulfilling
0:33:14 sort of compounding loop.
0:33:16 And then of course he’s got the Tesla fleet,
0:33:19 like Aaron was saying earlier with FSD.
0:33:21 I’m proud to have been a month one subscriber for it.
0:33:24 And I’m happy that I’m contributing to that training cycle
0:33:26 because it makes my Model X smarter next time around.
0:33:29 So the challenge then is for companies that don’t have
0:33:32 their own full stack, sort of fully integrated environment,
0:33:34 right, where they don’t have deployments out in the field.
0:33:37 And Darren’s point, you can either take the simulation route
0:33:38 for that and say we’re going to create
0:33:40 these sort of synthetic pipelines.
0:33:43 Or we’re seeing this huge build out of teleop fleets.
0:33:44 Like with language models,
0:33:46 you had people all around the world in countries
0:33:47 showing up and labeling data.
0:33:50 You have teleop fleets of people
0:33:52 piloting mechanical arms halfway around the world.
0:33:56 I think there’s an interesting sort of third new category
0:33:57 of efforts we’re tracking,
0:33:59 which is crowdsourced coalitions, right?
0:34:02 So an example of this is the DeepMind team
0:34:04 put out this maybe a year and a half ago,
0:34:06 robotics dataset called RTX,
0:34:08 where they partnered with a bunch of academic labs
0:34:09 and said, hey, you send us your data.
0:34:11 We’ve got compute and researchers.
0:34:13 We’ll train the model on your data and then send it back to you.
0:34:15 And what’s happening is there’s just different labs
0:34:18 around the world who have different robots of different kinds.
0:34:20 Some are arms, some are quadruped, some are bipeds.
0:34:21 And so instead of needing all of those
0:34:23 to be centralized in one place,
0:34:25 there’s a decoupling happening where some people are saying,
0:34:28 well, we’ll specialize in providing the compute
0:34:30 and the research talent, and then you guys bring the data.
0:34:32 And then it’s a give-to-get model, right?
0:34:34 Which we saw in some cases with the internet early on.
0:34:35 And Vidya is an example of this
0:34:39 where their research team is stacking a bunch of robots in-house.
0:34:41 So they instead partnering with people like pharma labs
0:34:44 who have arms doing the betting and wet lab experiments
0:34:45 and saying, you send us the data.
0:34:46 We’ve got a bunch of GPUs.
0:34:47 We’ve got some talented deep learning folks.
0:34:49 We’ll train the model, send it back to you.
0:34:51 And I think it’s an interesting experiment.
0:34:53 And there’s reason to believe this sort of give-to-get model
0:34:56 might end up actually having the highest diversity of data.
0:34:59 But we’re definitely in full experimentation land right now.
0:35:01 Yeah, and my guess is we’ll need all of it.
0:35:03 So it sounds like data is a big gap
0:35:05 and it sounds like some builders are working on that.
0:35:08 But where would you guys like to see more builders focused
0:35:11 in this hardware software arena,
0:35:14 especially because I do think there are some consumer-facing areas
0:35:15 where people are drawn to.
0:35:16 They see an event like this and they’re like,
0:35:18 oh, I want to work on that.
0:35:20 Yeah, I’m pretty excited about the long tail
0:35:23 of really unsexy industries
0:35:26 that have outsized impact on our GDP
0:35:28 and are often really critical industries
0:35:31 where people haven’t really been building for a while,
0:35:36 things like energy, manufacturing, supply chain, defense.
0:35:38 These industries that really carry the U.S. economy
0:35:41 and where we have under-invested from a technology perspective,
0:35:43 probably in the last several decades,
0:35:45 are poised to be pretty transformed
0:35:48 by this sort of hardware software melding in autonomy.
0:35:49 I’d love to see more people there.
0:35:52 I’m very excited for all the applications
0:35:53 they’re in and talked about.
0:35:54 And I think to unlock those,
0:35:57 we really need a way to solve this data bottleneck, right?
0:36:00 So startups, builders who are figuring out really novel ways
0:36:02 to collect that data in the world,
0:36:04 get it to researchers, make sense of it, curate it.
0:36:06 I think that’s sort of a fundamental limit
0:36:08 around progress across all of these industries.
0:36:09 We just need to sort of 10x
0:36:11 the rate of experimentation in that space.
0:36:16 All right, that is all for today.
0:36:19 If you did make it this far, first of all, thank you.
0:36:21 We put a lot of thought into each of these episodes,
0:36:23 whether it’s guests, the calendar tetras,
0:36:25 the cycles with our amazing editor, Tommy,
0:36:27 until the music is just right.
0:36:29 So if you like what we put together,
0:36:33 consider dropping us a line at ratethespodcast.com/a16z.
0:36:36 And let us know what your favorite episode is.
0:36:38 It’ll make my day, and I’m sure Tommy’s too.
0:36:40 We’ll catch you on the flip side.
0:36:43 (gentle music)
0:36:53 [BLANK_AUDIO]