The Exchanges

Q can be, obviously a large amount of AI is driven by large data sets, but I had a guest on the show the other day that went counter to this, and he said the value of massive data sets is largely overplayed. Frank, tell me, is that a fair assessment, and how do you think about the importance of massive data sets at scale in the effectiveness of AI implementation?

A Well, if you think about one of the most Popular machine learning techniques today, it's called deep learning, and it absolutely depends on large data sets. So if you think about deep learning in the context of self-driving cars, they're using deep learning to figure out, is that a stop sign? Is that a pedestrian? Is that a biker? And that absolutely depends on large data sets where you're taking the video and other The human labelers and asking them to identify what is actually in this frame of video. So there are certain tasks for which large data sets matter a lot. Now, some interesting approaches to achieving very good AI without large data sets. So approach one, let's have the machine generate the data sets themselves. And so if you look at AlphaGo, which is a system that learned to beat world champions at Go, they basically generated their own data sets by playing each other. So no human games involved. So that's one way to get around this big data set problem. Another is to use a set of techniques that does not depend on big data sets. And so reinforcement learning is an example of a type of machine learning that doesn't require a massive data set. It's sort of, you let the system play in a simulated environment and it sort of figures out what to do. So programming robots to put stuff in the boxes that Amazon is sending you sort of Take it back. Those types of things are…

AI assessment note: “So there are certain tasks for which large data sets matter a lot.”

Q terms of the car ownership element there. In terms of the ultimate owner, I'm too intrigued not to ask. Is that, you know, I'm always perplexed by kind of Tesla moving down the stack into having an Uber-style service, and then Uber moving up the stack into having a Tesla-style car creation process. How do you think about the ultimate owner of those cars in this kind of service model?

A Yeah, it's a little unclear how it will play out. It's certainly possible that the owners are the manufacturers, right? You mentioned Tesla, the BMW just offered, uh, started a ride hailing service in China. And so if you think about BMW, they're going to make the transition possibly from making the ultimate driving machine to operating the ultimate V driven service. That is a massive transition. I actually expect most of the manufacturers to not be able to make that transition, because if you think about every bonus that has ever been paid out at a car manufacturer, it's because you designed a car that people really wanted the three series. Or you worked with an agency to come up with the tagline ultimate driving machine, or you created the leasing model that drove your market share up five points, right? Like all of those things were aimed for getting a consumer to buy or lease your car to design a whole new set of incentives for your people to say, look, I want utilization, or I want you to be Uber, or all of those things is going to be monumentally difficult for companies that have been doing it Very, very well and very, very profitably for, in some cases, a hundred years. So I think the manufacturers are going to have a hard time making the transition. It's the ultimate innovator's dilemma. So if you think maybe the OEMs, the car manufacturers, aren't going to do it, it's …

AI assessment note: “It's certainly possible that the owners are the manufacturers”

Q I couldn't agree with you more, and I do have to ask, I studied attrition rates before across the world, and in the Valley, it's apparently 1.4 years average duration at a company today. Do you think that's a foundational problem that maybe the Valley lacks that? Maybe a more foundational loyalty towards the mission and the relationships that you have with the likes of Ben and Mark?

A You know, it is both a feature and a bug, as I pointed out. If you're looking at it as a feature, you think, look, we have a free-flowing set of ideas and people moving between companies, And actually academics who have studied why Silicon Valley beat route one, 28 in Boston, they will point to this as a feature, which is in Boston. You basically settled in a deck or Lotus, and you basically spent your entire career there. You might think of different ecosystems where sort of Japan and Korea, where this is still prize. Like my parents grew up in a generation where lifetime employment at the same company was the ideal and anything else that sort of fell short of that was considered suspicious. And so I That's sort of the feature point of view. The bug point of view on it is as a manager and as an entrepreneur trying to build a durable company that can weather the many, many storms you're going to go through as a startup. And so if you've got people leaving every 1.4 years, it's hard to keep people in those seats. It's hard to achieve the continuity. It's hard to keep the tribal knowledge in the system. And so that's sort of the bug point of view. And I think we need to strike the right balance. In building great companies, you're not going to hang on to everybody, but it would be good to be skilled enough and persuasive enough and have high enough to EQ to convince the people th…

AI assessment note: “it is both a feature and a bug, as I pointed out.”

Q I do want to dive into one of the dominant themes of kind of vertical AI being applied today, and we touched on it earlier with autonomous trucking, and it's kind of the autonomous car industry. I want to start on ownership and the foundation, so to speak. So how do you foresee the ownership structure and usage model playing out for the self-driving car industry over the coming years?

A Yeah, I think that in most urban and most suburban settings, people will stop owning their cars. So we've already seen the massive growth of companies like Lyft and Uber. Extension of that is why should I own a car at all? If you think about the biggest part of the cost structure of delivering rides as a service today, it is the driver. And so once we get the driver out of the equation, then it will be cheaper for you and I to be chauffeured everywhere than it will be to own and operate our own car. So to put that in dollar terms, most analysts would say it costs about a buck or a buck 50 a mile for you to own and operate your own car. So that's depreciation and insurance and fuel and maintenance and so on and so forth. Most of the self-driving estimates put the car, the cost of operating a profitable self-driving service at around 50 cents, so literally sort of half or a third of the price. I think even when we get to price parity, there's going to be a lot of people switching. When we get to, it's cheaper to be chauffeured

AI assessment note: “in most urban and most suburban settings, people will stop owning their cars.”

Q look at the kind of current landscape of levels of self-driving today, that we're currently in this kind of evolutionary phase, so to speak, being human input still being required, but, you know, automation still being possible. I'm intrigued, how do you think about this kind of Phase driven deployment process was self-driving. Do you think that's what we'll see, or will it be maybe more of a binary shift?

A So I'm thinking it's going to be more of a binary shift because the cars where you need human input, so you think of Tesla Autopilot, or you think of Nissan's ProPilot, right? These are products that are driver assist features. They put the driver in a very tough situation, which is you can sort of pay attention, and when you really need to pay attention, I'm going to sound a chime or something. I think that kind of half paying attention is very, very dangerous for humans, right? So we already see this with planes when autopilots are routinely engaged pilot skills atrophy, right? And so there's a well-documented history of plane accidents that are caused by pilots just forgetting how to fly the plane because they don't really need to most of the time. So I think from a skill atrophy point of view, and from an attention point of view, we're in this transition phase where all this driver assist Is going to get phased out in favor of complete driving. Now, on complete driving, I think it's going to unfold, not we wake up one day in a city and all the cars drive themselves. So first, we're going to do things like long-haul trucking, where there's a fixed route, and we can map all the routes in advance, and it's mostly freeway driving. And then it will be in geofenced areas. So think retirement communities, universities, military bases, where in general, we're So I think in those ar…

AI assessment note: “So I'm thinking it's going to be more of a binary shift because”

Q I do want to dive into one of the dominant themes of kind of vertical AI being applied today, and we touched on it earlier with autonomous trucking, and it's kind of the autonomous car industry. I want to start on ownership and the foundation, so to speak. So how do you foresee the ownership structure and usage model playing out for the self-driving car industry over the coming years?

A Yeah, I think that in most urban and most suburban settings, people will stop owning their cars. So we've already seen the massive growth of companies like Lyft and Uber. Extension of that is why should I own a car at all? If you think about the biggest part of the cost structure of delivering rides as a service today, it is the driver. And so once we get the driver out of the equation, then it will be cheaper for you and I to be chauffeured everywhere than it will be to own and operate our own car. So to put that in dollar terms, most analysts would say it costs about a buck or a buck 50 a mile for you to own and operate your own car. So that's depreciation and insurance and fuel and maintenance and so on and so forth. Most of the self-driving estimates put the car, the cost of operating a profitable self-driving service at around 50 cents, so literally sort of half or a third of the price. I think even when we get to price parity, there's going to be a lot of people switching. When we get to, it's cheaper to be chauffeured

AI assessment note: “in most urban and most suburban settings, people will stop owning their cars.”

Q No, I love that kind of the best that we've ever been. I do have to ask, maybe from a more financially macro perspective, if we think about the best humans in the society that we inhabit, how does that maybe shift the economy? I'm always really intrigued by kind of automation's role on class distinction. Does it maybe change class distinctions?

A I'm not sure it changes class distinctions, but the way I think about it is that there's a set of things that machines are fantastic at. And so think about sort of the spreadsheet error or the database error. It was just true that spreadsheets and databases remembered more, could do math faster, could do more math more accurately, and it turbocharged humans. And where we were actually afraid when the spreadsheet came out that, oh my God, well, there go all the white collar jobs. We've actually created more white collar jobs post the spreadsheet than we have in the decades before. And so, it didn't turn out to wipe out all the jobs. It turbocharged us, and it created new jobs. So, AI, I think, is going to be like that, which is, there's a bunch of things that AIs can now do that spreadsheets and databases could never do. It will turbocharge our capabilities. It will do some of the most dangerous and dirty jobs. So, think about power plant inspections. Think about cell tower inspections. Think about improving mine safety. One of the very little-known facts is that one of the most dangerous jobs on the The planet is long haul trucking, which is in the United States. This causes the most fatalities being a long haul trucker. So look, we've got to get the truckers out of the cabs as soon as we can. If we can invent a technology that can drive long haul trucks safely. And you know, t…

AI assessment note: “I'm not sure it changes class distinctions, but the way I think about it”

Q look at the kind of current landscape of levels of self-driving today, that we're currently in this kind of evolutionary phase, so to speak, being human input still being required, but, you know, automation still being possible. I'm intrigued, how do you think about this kind of Phase driven deployment process was self-driving. Do you think that's what we'll see, or will it be maybe more of a binary shift?

A So I'm thinking it's going to be more of a binary shift because the cars where you need human input, so you think of Tesla Autopilot, or you think of Nissan's ProPilot, right? These are products that are driver assist features. They put the driver in a very tough situation, which is you can sort of pay attention, and when you really need to pay attention, I'm going to sound a chime or something. I think that kind of half paying attention is very, very dangerous for humans, right? So we already see this with planes when autopilots are routinely engaged pilot skills atrophy, right? And so there's a well-documented history of plane accidents that are caused by pilots just forgetting how to fly the plane because they don't really need to most of the time. So I think from a skill atrophy point of view, and from an attention point of view, we're in this transition phase where all this driver assist Is going to get phased out in favor of complete driving. Now, on complete driving, I think it's going to unfold, not we wake up one day in a city and all the cars drive themselves. So first, we're going to do things like long-haul trucking, where there's a fixed route, and we can map all the routes in advance, and it's mostly freeway driving. And then it will be in geofenced areas. So think retirement communities, universities, military bases, where in general, we're So I think in those ar…

AI assessment note: “So I'm thinking it's going to be more of a binary shift because”

Q perhaps the most meaty element, I speak to many different experts on the topic, and some suggest AI's ability for human augmentation and the benefits, and others suggest we've never seen the rate of adoption like this before, including horse and carts. And so we're in a little bit of a sticky situation. Where do you sit on the pendulum, so to speak, and what's that current state of play?

A Yeah, well, let me share a couple of factoids first that sort of ground what's happening in this ecosystem, and they're fun. So first is the biggest AI conference for researchers is called the Neuroinformation Processing Symposium. This year, it's sold out in a little over 11 minutes. It's like a Beyonce concert. So that's research. The second is, if you talk to Paul Doherty, who is the CTO of Accenture, and he's been at Accenture for three decades, he says he's never seen anything grow this fast inside big businesses, right? So if you think about headcount, think about budget, like any metric you can think of, it's faster than the rise of client-server computing. It's faster than the move to internet computing. It's faster than shifting to mobile. This is the big one so far as he's seen in 30 years at Accenture. So that's pretty impressive. And then you know the It's big in tech when all the politicians are waving their flag. And in my country, you will be rewarded. There'll be tax breaks. There'll be regions. There'll be collaborations with universities, and I will be the best country or region on the planet to bring your AI researcher to build an AI powered startup. And so look, you have researchers, you have the business community, you have politicians waiting the flag. So that just gives you a sense of the tremendous excitement that's happening. And so I think it's up to u…

AI assessment note: “we are going to be the best humans we have ever been, period.”

Q can be, obviously a large amount of AI is driven by large data sets, but I had a guest on the show the other day that went counter to this, and he said the value of massive data sets is largely overplayed. Frank, tell me, is that a fair assessment, and how do you think about the importance of massive data sets at scale in the effectiveness of AI implementation?

A Well, if you think about one of the most Popular machine learning techniques today, it's called deep learning, and it absolutely depends on large data sets. So if you think about deep learning in the context of self-driving cars, they're using deep learning to figure out, is that a stop sign? Is that a pedestrian? Is that a biker? And that absolutely depends on large data sets where you're taking the video and other The human labelers and asking them to identify what is actually in this frame of video. So there are certain tasks for which large data sets matter a lot. Now, some interesting approaches to achieving very good AI without large data sets. So approach one, let's have the machine generate the data sets themselves. And so if you look at AlphaGo, which is a system that learned to beat world champions at Go, they basically generated their own data sets by playing each other. So no human games involved. So that's one way to get around this big data set problem. Another is to use a set of techniques that does not depend on big data sets. And so reinforcement learning is an example of a type of machine learning that doesn't require a massive data set. It's sort of, you let the system play in a simulated environment and it sort of figures out what to do. So programming robots to put stuff in the boxes that Amazon is sending you sort of Take it back. Those types of things are…

AI assessment note: “So there are certain tasks for which large data sets matter a lot.”

Q Aaron Van De Vende from Founders Fund on the show recently, actually, and he stated, the majority of AI is a scam, and it's largely a case of rebranding in most circumstances, be it from statistics to big data to every other nomenclature that we've gone through over the years. Is that fair from your perspective? And how would you respond to that kind of merely a change of nomenclature?

A I'm not sure I would say all AI is a scam. I will say, I think I understand the impulse behind that statement, which is about three years ago, as we looked at startups, everybody became an AI startup. They got the .ai domain name. It was the second phrase in their pitch decks, which is, I mean, AI powered such and such. And look, some of those were real applications of AI, and some of them was sort of marketing a veneer over things that they'd always been doing, like, You know, data analysis. On the balance, though, I think there are genuinely new techniques that big companies and small companies are putting into their software that make them better and more useful. The types of predictions we can make, like, do you have cancer or not? The types of things that we can do, like, is your grandmother in this picture? The types of questions that we can answer, like, where should the soccer players be on the soccer field? Those were things that we could never do with the previous generation of technologies. No matter how hard you asked an Oracle database, Where should the soccer players be? You were never going to get a good answer. And so there are really new questions that we can ask and answer of our software that we've never been able to do before. My partner, Benedict Evans uses a great analogy of imagine that you had a million interns and the million interns could look at all o…

AI assessment note: “On the balance, though, I think there are genuinely new techniques”

Q terms of the car ownership element there. In terms of the ultimate owner, I'm too intrigued not to ask. Is that, you know, I'm always perplexed by kind of Tesla moving down the stack into having an Uber-style service, and then Uber moving up the stack into having a Tesla-style car creation process. How do you think about the ultimate owner of those cars in this kind of service model?

A Yeah, it's a little unclear how it will play out. It's certainly possible that the owners are the manufacturers, right? You mentioned Tesla, the BMW just offered, uh, started a ride hailing service in China. And so if you think about BMW, they're going to make the transition possibly from making the ultimate driving machine to operating the ultimate V driven service. That is a massive transition. I actually expect most of the manufacturers to not be able to make that transition, because if you think about every bonus that has ever been paid out at a car manufacturer, it's because you designed a car that people really wanted the three series. Or you worked with an agency to come up with the tagline ultimate driving machine, or you created the leasing model that drove your market share up five points, right? Like all of those things were aimed for getting a consumer to buy or lease your car to design a whole new set of incentives for your people to say, look, I want utilization, or I want you to be Uber, or all of those things is going to be monumentally difficult for companies that have been doing it Very, very well and very, very profitably for, in some cases, a hundred years. So I think the manufacturers are going to have a hard time making the transition. It's the ultimate innovator's dilemma. So if you think maybe the OEMs, the car manufacturers, aren't going to do it, it's …

AI assessment note: “It's possible that their ultimate ownership is by leasing companies.”

Q but the final element that I do want to touch on with regards to the kind of autonomous car space itself is really more the parts themselves, uh, integrating software into the equation. Today, the US spends a trillion dollars per year on car parts and servicing. Gus, how does this change with the rise of self-driving, and how do we think about that kind of fundamental supply chain shift?

A Yeah, so the biggest shift that will happen in the supply chain is think of the percentage of the total cost of the car It goes to hardware and software. So electronics and software. So today that's anywhere between, you know, sort of 10 and 20, 30%, depending on the, on the car you drive and how many fancy features it has. Like, you know, can it park itself? Does it have a lot of these so-called ADAS features that can drive on the freeway? So, you know, the more sophisticated the car, the more electronics and software. I think you're going to see, you know, half the price of a car go to hardware and software. So That's a boon, obviously, for people who are making those components, and it comes at the expense of people who are making older style components. So beginning with the internal combustion engine, think about your steering wheel, think about the exhaust pipe, think about your rear view mirrors, right? All of those will go away because the self-driving cars don't need rear view mirrors. They don't need steering wheels. They don't need brake pedals. They'll be doing all that through software.

AI assessment note: “the biggest shift that will happen in the supply chain is think of the percentage”

Q I'm just really intrigued when you think about kind of fundamentally reshaping our existing cities. How do we think about that when they are so established and so concrete in their structure? How do we think about the ease of transitioning towards an autonomous kind of structured city itself? Much easier if it was a new city from afresh.

A Yeah, there's an awful lot of Hardware, right? Concrete, asphalt, bridges, and so on. And then there is also a decidedly non-software governance process, right? Which is civil planning mediated by governments, right? So this is not going to happen fast, the redesign of our cities. So it's going to happen when parking lot One car dealership at a time. And we're going to make these local decisions like, gee, if we don't own cars, then we don't need to sell them. Therefore, car dealerships are going to start vanishing from inner cities. What do we do with that space? What do we do with the DMVs? I think the UK calls them DVLAs, right? So these places where we go and get our licenses, well, what do we do with that space? Because we don't need licenses anymore. We're being chauffeured. And so as these buildings disappear from the infrastructure, we get to think about How do we recycle and reuse for a better and higher purpose in each city? And of course, each city will have a different answer to that question, depending on what their economy and what their social goals are for the city.

AI assessment note: “So this is not going to happen fast... going to happen one parking lot”

Q I'm sure you do, but I would love to get the ball rolling today, Frank, by discussing a little bit about you. So tell me, how did you make your way into the world of startups and venture and come to be a partner today at Andreessen?

A Well, I kind of did it by accident, which is unlike many of my peers, I had not plotted a lifelong path to get to venture. I kind of fell into it backwards because Mark and Ben, sort of the Andreessen and the Horowitz of Andreessen Horowitz were doing it, and I have been working with these guys now for 25 odd years, and they're always doing interesting things, and I'm like, hey, if they're going to start a venture firm, maybe I'll go see what that's like, and the thing that has been awesome for me is I have a super curious personality. I'm always, like, The four-year-old asking, why? Why is that? Why is the sky blue? Why do, why does gravity pull us down? And for somebody with that kind of innate curiosity, this is the best job ever, right? Because you're going to learn something new every hour if you do your job right, and that's exactly the situation I'm in.

AI assessment note: “I kind of fell into it backwards because Mark and Ben”

Q I couldn't agree with you more, and I do have to ask, I studied attrition rates before across the world, and in the Valley, it's apparently 1.4 years average duration at a company today. Do you think that's a foundational problem that maybe the Valley lacks that? Maybe a more foundational loyalty towards the mission and the relationships that you have with the likes of Ben and Mark?

A You know, it is both a feature and a bug, as I pointed out. If you're looking at it as a feature, you think, look, we have a free-flowing set of ideas and people moving between companies, And actually academics who have studied why Silicon Valley beat route one, 28 in Boston, they will point to this as a feature, which is in Boston. You basically settled in a deck or Lotus, and you basically spent your entire career there. You might think of different ecosystems where sort of Japan and Korea, where this is still prize. Like my parents grew up in a generation where lifetime employment at the same company was the ideal and anything else that sort of fell short of that was considered suspicious. And so I That's sort of the feature point of view. The bug point of view on it is as a manager and as an entrepreneur trying to build a durable company that can weather the many, many storms you're going to go through as a startup. And so if you've got people leaving every 1.4 years, it's hard to keep people in those seats. It's hard to achieve the continuity. It's hard to keep the tribal knowledge in the system. And so that's sort of the bug point of view. And I think we need to strike the right balance. In building great companies, you're not going to hang on to everybody, but it would be good to be skilled enough and persuasive enough and have high enough to EQ to convince the people th…

AI assessment note: “it is both a feature and a bug”

Q What would you most like to change about the world of technology and venture itself?

A The thing I'd like to most change is to bring a lot more transparency, especially to the interaction between investors and entrepreneurs. I feel like we've gone into a phase where we need the ecosystem encourages hiding stuff from each other. And so the metrics in the pitch decks are hiding key facts. And I think, look, We're in this journey together. We want to help you create a successful business. And so beginning of that conversation by hiding key facts about your business is not helpful. And then on our side as investors, I think we can be a lot more transparent. We're certainly trying an A-sixteen Z in how we make decisions and what we're looking for in companies. Let's not hide things from each other. Let's get started on transparency and mutual trust.

AI assessment note: “The thing I'd like to most change is to bring a lot more transparency”

Q but the final element that I do want to touch on with regards to the kind of autonomous car space itself is really more the parts themselves, uh, integrating software into the equation. Today, the US spends a trillion dollars per year on car parts and servicing. Gus, how does this change with the rise of self-driving, and how do we think about that kind of fundamental supply chain shift?

A Yeah, so the biggest shift that will happen in the supply chain is think of the percentage of the total cost of the car It goes to hardware and software. So electronics and software. So today that's anywhere between, you know, sort of 10 and 20, 30%, depending on the, on the car you drive and how many fancy features it has. Like, you know, can it park itself? Does it have a lot of these so-called ADAS features that can drive on the freeway? So, you know, the more sophisticated the car, the more electronics and software. I think you're going to see, you know, half the price of a car go to hardware and software. So That's a boon, obviously, for people who are making those components, and it comes at the expense of people who are making older style components. So beginning with the internal combustion engine, think about your steering wheel, think about the exhaust pipe, think about your rear view mirrors, right? All of those will go away because the self-driving cars don't need rear view mirrors. They don't need steering wheels. They don't need brake pedals. They'll be doing all that through software.

AI assessment note: “the biggest shift that will happen in the supply chain is think of the percentage”

Q have to ask you, you mentioned the element of working with both Mark and Ben for a considerable time, and obviously the hard thing about hard things is such a hailed book in our industry with the lessons from Ben from Opsware. You also experienced that journey firsthand, so I have to ask you, Frank, and super interested by your perspective, what were some of the core takeaways for you?

A Yeah, so two big things from my experience, and this is sort of Netscape and Loud Cloud and Optware, which is, one, it is a lot harder to build a business than it looks from the outside, and when you read the hard thing about hard things, you sort of realize, yeah, boy, that thing, we nearly died a hundred times, and then we had lunch every day, you know, so it is a lot harder to build a big Durable business than it looks from the outside. And then two, the thing that I really learned was the value of loyalty, because when the company is not doing well, what happens as a manager and a leader in the company is you spend all of your time basically talking people from leaving the company. And if you think about the ecosystem that you're in, there's always a shinier new thing in Silicon Valley. There's always a Google or a Facebook or a Some other cool place to go, and there's a natural flow of talent. You know, some would argue, look, this is one of the things that keeps the valley what it is, which is there's a free flow of people and information and ideas. Which is from the point of view of the company or a manager, like having people leave every 18 months, that's not good for your company. And so I learned a lot about how to inspire loyalty and certainly taught me a lot about staying in place when things didn't look like they were going well, because it's hard enough building a…

AI assessment note: “two big things from my experience, and this is sort of Netscape and Loud Cloud”

Q I do want to finish, and before we move into the quick On the cities that these cars really drive in, starting on, again, the foundations being kind of the public infrastructure layer, how will both the rise of automation and autonomous cars change? Maybe how we approach, build, and use public infrastructure in the future?

A Well, the super exciting thing is we've really only had three chapters of city design in the history of cities. So chapter one was we walked around in our cities, and then chapter two was the horses came. And then chapter three was, okay, the internal combustion engine powered cars are here. And then you could argue there was sort of a chapter three B, which is the freeway Now you count them three or four chapters. When we have self-driving cars, we'll get to design the next chapter, and if we think about designing the next chapter, we can completely rethink how we use the space in our cities. So first, all the parking lots can go away because the cars will constantly be in motion. The fleet operators will need that to drive the efficiency and utilization out of their cars, and so if the car is not carrying you, it's carrying goods, and if it's not carrying a good, then it's On its way to being cleaned, or serviced, or charged, right? So there will be no need for parking lots, and if we do need parking lots, because we do need to plan for peak utilization in a city, we're not going to put them in a city. Like, if you were the Sin City designer of a city, you'd put that in a suburb somewhere, where the cars can go and rest overnight. Like, imagine that you could walk into the city planning office of Los Angeles and say, hello, today, 14% of all land in Los Angeles is parking lot…

AI assessment note: “all the parking lots can go away because the cars will constantly be in motion.”

Q have to ask you, you mentioned the element of working with both Mark and Ben for a considerable time, and obviously the hard thing about hard things is such a hailed book in our industry with the lessons from Ben from Opsware. You also experienced that journey firsthand, so I have to ask you, Frank, and super interested by your perspective, what were some of the core takeaways for you?

A Yeah, so two big things from my experience, and this is sort of Netscape and Loud Cloud and Optware, which is, one, it is a lot harder to build a business than it looks from the outside, and when you read the hard thing about hard things, you sort of realize, yeah, boy, that thing, we nearly died a hundred times, and then we had lunch every day, you know, so it is a lot harder to build a big Durable business than it looks from the outside. And then two, the thing that I really learned was the value of loyalty, because when the company is not doing well, what happens as a manager and a leader in the company is you spend all of your time basically talking people from leaving the company. And if you think about the ecosystem that you're in, there's always a shinier new thing in Silicon Valley. There's always a Google or a Facebook or a Some other cool place to go, and there's a natural flow of talent. You know, some would argue, look, this is one of the things that keeps the valley what it is, which is there's a free flow of people and information and ideas. Which is from the point of view of the company or a manager, like having people leave every 18 months, that's not good for your company. And so I learned a lot about how to inspire loyalty and certainly taught me a lot about staying in place when things didn't look like they were going well, because it's hard enough building a…

AI assessment note: “two big things from my experience, and this is sort of Netscape and Loud Cloud”

Q I'm sure you do, but I would love to get the ball rolling today, Frank, by discussing a little bit about you. So tell me, how did you make your way into the world of startups and venture and come to be a partner today at Andreessen?

A Well, I kind of did it by accident, which is unlike many of my peers, I had not plotted a lifelong path to get to venture. I kind of fell into it backwards because Mark and Ben, sort of the Andreessen and the Horowitz of Andreessen Horowitz were doing it, and I have been working with these guys now for 25 odd years, and they're always doing interesting things, and I'm like, hey, if they're going to start a venture firm, maybe I'll go see what that's like, and the thing that has been awesome for me is I have a super curious personality. I'm always, like, The four-year-old asking, why? Why is that? Why is the sky blue? Why do, why does gravity pull us down? And for somebody with that kind of innate curiosity, this is the best job ever, right? Because you're going to learn something new every hour if you do your job right, and that's exactly the situation I'm in.

AI assessment note: “I kind of fell into it backwards because Mark and Ben... were doing it”

Q perhaps the most meaty element, I speak to many different experts on the topic, and some suggest AI's ability for human augmentation and the benefits, and others suggest we've never seen the rate of adoption like this before, including horse and carts. And so we're in a little bit of a sticky situation. Where do you sit on the pendulum, so to speak, and what's that current state of play?

A Yeah, well, let me share a couple of factoids first that sort of ground what's happening in this ecosystem, and they're fun. So first is the biggest AI conference for researchers is called the Neuroinformation Processing Symposium. This year, it's sold out in a little over 11 minutes. It's like a Beyonce concert. So that's research. The second is, if you talk to Paul Doherty, who is the CTO of Accenture, and he's been at Accenture for three decades, he says he's never seen anything grow this fast inside big businesses, right? So if you think about headcount, think about budget, like any metric you can think of, it's faster than the rise of client-server computing. It's faster than the move to internet computing. It's faster than shifting to mobile. This is the big one so far as he's seen in 30 years at Accenture. So that's pretty impressive. And then you know the It's big in tech when all the politicians are waving their flag. And in my country, you will be rewarded. There'll be tax breaks. There'll be regions. There'll be collaborations with universities, and I will be the best country or region on the planet to bring your AI researcher to build an AI powered startup. And so look, you have researchers, you have the business community, you have politicians waiting the flag. So that just gives you a sense of the tremendous excitement that's happening. And so I think it's up to u…

AI assessment note: “we are going to be the best humans we have ever been”

Q No, I love that kind of the best that we've ever been. I do have to ask, maybe from a more financially macro perspective, if we think about the best humans in the society that we inhabit, how does that maybe shift the economy? I'm always really intrigued by kind of automation's role on class distinction. Does it maybe change class distinctions?

A I'm not sure it changes class distinctions, but the way I think about it is that there's a set of things that machines are fantastic at. And so think about sort of the spreadsheet error or the database error. It was just true that spreadsheets and databases remembered more, could do math faster, could do more math more accurately, and it turbocharged humans. And where we were actually afraid when the spreadsheet came out that, oh my God, well, there go all the white collar jobs. We've actually created more white collar jobs post the spreadsheet than we have in the decades before. And so, it didn't turn out to wipe out all the jobs. It turbocharged us, and it created new jobs. So, AI, I think, is going to be like that, which is, there's a bunch of things that AIs can now do that spreadsheets and databases could never do. It will turbocharge our capabilities. It will do some of the most dangerous and dirty jobs. So, think about power plant inspections. Think about cell tower inspections. Think about improving mine safety. One of the very little-known facts is that one of the most dangerous jobs on the The planet is long haul trucking, which is in the United States. This causes the most fatalities being a long haul trucker. So look, we've got to get the truckers out of the cabs as soon as we can. If we can invent a technology that can drive long haul trucks safely. And you know, t…

AI assessment note: “I'm not sure it changes class distinctions, but the way I think about it”