The Exchanges

Q Well, there was no tiny minute VC at the time, was there? So I don't blame you.

A Yeah, exactly. I mean, the world was so different. Even when I graduated from college, I mean, there were five people in my class that were interested in startups. And I met with a few venture firms. I always had this idea that I would go out and start a company, but I thought a venture firm would be a good place To learn a little bit more and kind of recharge a little bit. And I fortuitously met the folks at General Catalyst and just really resonated with the team. So I joined at a great time in the industry in 2010 with General Catalyst, focused on early stage investing in areas like SMB software, small business software, the on-demand wave, but more broadly, just thinking about amazing founders, especially young technical founders. When I was there, I was lucky to co-found this initiative called Rough Drop Ventures, which is an initiative that's a Support student entrepreneurs, which was a real passion of mine and continues to be a real interest. And I was very lucky to be invested in some really tremendous companies as an investor at GC. I told myself if I found the right person and idea that I thought was big enough, I really would love to try and go out and start a company. And so even though I spent more years in venture than I thought I would, things started to come together when I started to chat with my co-founder, Adam, and start to get really fascinated with Things …

AI assessment note: “Yeah, exactly. I mean, the world was so different.”

Q Your favorite blog or newsletter? What are your must reads?

A I read all the standard tech news outlets, like, TechCrunch and everything else just to stay on top of news. I feel like Bill Gurley probably creates some of the highest quality individual posts. I'm always curious when he, when he produces something. Other than that, I don't think I'm the most loyal reader in the world. I use Nuzzle, Twitter, Medium to find new articles. I have become a big podcast person over the last few months, so, you know, check out the show a ton, the A-sixteen Z podcast, Recode. There's a bunch of great content out there, but I tend to surface it through networks versus lowly going to any one thing, except for the show, of course.

AI assessment note: “Bill Gurley probably creates some of the highest quality individual posts.”

Q machine there, are you bullish on individual teams creating kind of proprietary and, and their own different algorithms, or whether we'll see your IBMs creating machine learning as a service, and that will really become a real offering? So what do you think? Do you think we'll see this kind of monopoly of machine learning as a service, or do you think we'll see individual teams Tailoring their own algorithms.

A Yes, I think it's certainly true that the big companies have aggregated massive amounts of machine learning and machine learning research capabilities, and one of the really nice things is that a lot of these companies are open sourcing and democratizing these services, so things like TensorFlow from Google, and the result is, you know, we, when we recruit engineers, I mean, even an out-of-school engineer can do something now that only specialists could have done a few years ago, so I feel like There is definitely a democratization happening around some of the specialized machine learning services, which I think is a really great thing for the ecosystem because it enables startups and companies to create real applications. There probably are some startups that will do breakthrough machine learning, research, and intelligence, but I see a lot of them getting scooped up by big companies right now, and it's, I think, very hard to compete against all the resource and horsepower that these big companies are putting into research, and the fact that they're willing to democratize and open source these types of tools. One of the areas where I feel like startups Can have a distinct advantage is creating unique data sets that do interesting things around on top of some of these machine learning algorithms. And so, for example, with our service at B-Twelve, one of the first applications w…

AI assessment note: “very hard to compete against all the resource and horsepower that these big companies”

Q putting a figure on it, let's put figures on it before we dive into a quickfire. Let's do ANI until it's fully proliferated into the market without sticky tape. How long will that take? AGI and then ASI. So artificial general intelligence and then artificial super intelligence. Let's put a timeline on it. I'm going to be really annoying here. Let's put some bets down. So ANI fully proliferated when?

A Well, so I think you already see instances where we're firmly in the ANI phase. You see lots of instances where things are self-sufficient. And I think this is a really exciting phase because there are lots of narrow things like driverless cars, as an example, or other areas that are going to change the way that we work and live together. So I feel like we're firmly in that, in that space right now. And I think this phase will continue for, in my eyes, decades. So I think there's lots of opportunity for startups to continue building and operating in the space right now. If I were to think about AGI, the most optimistic people, the Ray Kurzweils of the world and others, put a time frame of 20 thirties, 20 forties as being the early, earliest instances of when that type of technology could exist. I feel like that would be on the very early end. And, you know, we might still be hundreds of years away from Ultimately creating machines that are broadly as smart as people. I actually don't think the limitation will be compute power, but the limitation will more be around the type of software that you need to build something as complex and dynamic as a person. And then I'm firmly a believer in exponential growth in technology. So I feel like once you get to AGI and you're able to have machines that can operate 24 seven, sufficiently work together and do further research in the space, …

AI assessment note: “ANI phase... will continue for, in my eyes, decades”

Q Now, I'm super intrigued. As a young person in venture at the time, did you find it quite a lonely place? Often for the juniors in venture, it's often spent DDing, researching, and spending a lot of time on your own. Did you find this to be challenging or a lonely place?

A Not really. I think at GC, there's a really great cohort of associates and principals, and so we had a really great group of people to work with. It is a little bit more of a solo sport, because anytime you're doing a deal, you're the person at the firm that's primarily responsible for it. I think compared to when I started in 2010 to when I left in 2015, there's just been an explosion of young people in the venture industry, and people are doing really incredible things. So it's very different than a startup Or a bigger company where, you know, day in and day out work with a team. But there's luckily a great support network, and the folks at GC have been, you know, very close mentors, friends, and partners through the whole process. So I think there was a lot of people to kind of communicate with. And then above and beyond that, you're spending so much time meeting with entrepreneurs, supporting other folks in the community. So you end up spending a lot of time meeting with people and interacting with people. So it was, it was always something that kind of like made the, the Process, and even some of the more analytic stuff, analytics-driven work, feel human, because there was a lot of personal interaction.

AI assessment note: “Not really. I think at GC, there's a really great cohort”

Q In terms of algorithm training there, how do you approach that with B-Twelve?

A So I think the way that we often think about things is before you start out and build out really fancy technology and tooling and algorithms, We want to really make sure we internalize workflows ourselves. So with our first product, the first thing we did is just really understand how exactly the website and web design process worked. We did things on our own, and just were able to create very simple rules around how we wanted things to work from a machine output perspective. And then as we start to get more and more scale, we're able to use a network of experts that we have to think about ways to get very specialized data sets specific to web design. And so a lot of it is, I think first is understanding the end output that you're trying to get to, and then thinking about how do you collect data from actual human and person usage that can make the experience better.

AI assessment note: “use a network of experts that we have to think about ways to get very specialized data sets”

Q putting a figure on it, let's put figures on it before we dive into a quickfire. Let's do ANI until it's fully proliferated into the market without sticky tape. How long will that take? AGI and then ASI. So artificial general intelligence and then artificial super intelligence. Let's put a timeline on it. I'm going to be really annoying here. Let's put some bets down. So ANI fully proliferated when?

A Well, so I think you already see instances where we're firmly in the ANI phase. You see lots of instances where things are self-sufficient. And I think this is a really exciting phase because there are lots of narrow things like driverless cars, as an example, or other areas that are going to change the way that we work and live together. So I feel like we're firmly in that, in that space right now. And I think this phase will continue for, in my eyes, decades. So I think there's lots of opportunity for startups to continue building and operating in the space right now. If I were to think about AGI, the most optimistic people, the Ray Kurzweils of the world and others, put a time frame of 20 thirties, 20 forties as being the early, earliest instances of when that type of technology could exist. I feel like that would be on the very early end. And, you know, we might still be hundreds of years away from Ultimately creating machines that are broadly as smart as people. I actually don't think the limitation will be compute power, but the limitation will more be around the type of software that you need to build something as complex and dynamic as a person. And then I'm firmly a believer in exponential growth in technology. So I feel like once you get to AGI and you're able to have machines that can operate 24 seven, sufficiently work together and do further research in the space, …

AI assessment note: “this phase will continue for, in my eyes, decades.”

Q You know, you were carving out a real name for yourself in the industry, and as you said, made some great investments. Was it difficult to leave the security of venture for the, for the risk of a startup?

A It gets harder with time. Economically, it gets harder. You get comfortable. You start to feel like you can actually do a good job in the venture industry, but it's just something I had this Itch ever since I was junior in college to go out and try and build a notable company. And so even as I was sitting on pitches, I knew that it was something that I'd want to do and I'm not doing. And I think just given the fact that I was able to team up with my co-founder, who was like the number one person I wanted to go and start a company with, and I felt like the opportunity was right. It felt really easy. I haven't looked back, you know, I think looking back at these career changes can be dangerous and the startup journey itself has been so much fun. But there's definitely a lot of considerations, and I think the, one of the things about doing venture when you're so early in your career is that it's a lot more nimble. I felt like there was always a path back in if that was the right move for me, but the opportunity in starting a company, especially in the space, felt a lot more unique at the time.

AI assessment note: “It felt really easy. I haven't looked back”

Q How much of narrow AI do you think is being held together by, by sticky tape, so to speak? For example, x.ai still has a lot of people checking the appointments that are being set by Amy, referring to potential sticky tape. How, how many of these narrow services do you think is still being held together slightly by sticky tape?

A Yeah, I feel like a lot of companies are able to To get to 80, 90, 95% automation. But the challenge is, like, most people don't care about, you know, if you could get a driverless car that could drive well 85% of the time, that's not particularly useful because that's still too dangerous or inconsistent experience. I think it's really going from like 90 to 95 to 99 to a hundred percent is where a lot of the work is. And in certain areas where we've bitten off experiences or want to use AI, it's unclear whether you can ultimately get there, which is why At B-Twelve, we're going after kind of what we call human-assisted AI, where we always think people will be in the loop. And as there's more technology, more capabilities, more compute power, we'll figure out ways in which automation can play a role. But it doesn't necessarily require us to a hundred percent automate the complex workflows that we go after. And I think each startup is different. In a lot of cases, there are people behind the machine that are masked or not put at the forefront. And in In certain cases, like, those people aren't going anywhere anytime soon if you want to create the type of experience I think consumers want.

AI assessment note: “In a lot of cases, there are people behind the machine that are masked”

Q What's your biggest concern for AI going forward? So mine, for example, is Tesla's going wrong and it creating mass public concern and a consensus that kind of robots are bad after all.

A Yeah, I think in the short term, my biggest concern is that the field over promises and under delivers, just because we're, it's so easy to make that jump. And I think a lot of people make claims that are just hard to substantiate right now. And as a result, a lot of the fantasy and enthusiasm has brought people into this space. It starts to fizzle away, which I think would be disappointing given all the opportunity there is here. Longer term, I'm probably most concerned about, are we able to build the right type of humanity into the various AIs that we build? And there's a lot of hard decisions to be made there, and as we start to get more sophisticated in the tooling that we build, can we really make them function and work in a way that aren't just optimizing around results, but optimizing in the, in the human way that all of us like to live and work and operate.

AI assessment note: “in the short term, my biggest concern is that the field over promises and under delivers”

Q So would you say the incumbency advantage with regards to data set is offset through the further Specialization of data sets, as you said there, into actually how users behave on websites. I don't know, heat maps as to where they click potentially. Do you know what I mean? Is it the further niche that we go is where the startups will offset the incumbency data advantage?

A I think it really depends on the particular space or application, but it certainly is true that a lot of the data that bigger companies have captured or will capture may or may not be the types of things that actually create the product experiences that startups Build or technology companies build and that consumers ultimately love. So I think there is an ability to do interesting things around creating proprietary and unique data sets that are way more applicable to the types of products that we want to build. In a lot of cases, collecting that data can be quite messy. You know, you use actual people to train an algorithm or use actual people to collect very unique and specific data. And whether you're a big company or a small company, that can be challenging. And often as a small company, it's a little bit easier because you don't have The reputation risk and or the challenges that a bigger company might have around organizing a cohort of people that in their world is very small, but in a startup world could be, you know, a very important group of people to take care of and treat well.

AI assessment note: “creating proprietary and unique data sets that are way more applicable”

Q You know, you were carving out a real name for yourself in the industry, and as you said, made some great investments. Was it difficult to leave the security of venture for the, for the risk of a startup?

A It gets harder with time. Economically, it gets harder. You get comfortable. You start to feel like you can actually do a good job in the venture industry, but it's just something I had this Itch ever since I was junior in college to go out and try and build a notable company. And so even as I was sitting on pitches, I knew that it was something that I'd want to do and I'm not doing. And I think just given the fact that I was able to team up with my co-founder, who was like the number one person I wanted to go and start a company with, and I felt like the opportunity was right. It felt really easy. I haven't looked back, you know, I think looking back at these career changes can be dangerous and the startup journey itself has been so much fun. But there's definitely a lot of considerations, and I think the, one of the things about doing venture when you're so early in your career is that it's a lot more nimble. I felt like there was always a path back in if that was the right move for me, but the opportunity in starting a company, especially in the space, felt a lot more unique at the time.

AI assessment note: “It felt really easy. I haven't looked back”

Q How much of narrow AI do you think is being held together by, by sticky tape, so to speak? For example, x.ai still has a lot of people checking the appointments that are being set by Amy, referring to potential sticky tape. How, how many of these narrow services do you think is still being held together slightly by sticky tape?

A Yeah, I feel like a lot of companies are able to To get to 80, 90, 95% automation. But the challenge is, like, most people don't care about, you know, if you could get a driverless car that could drive well 85% of the time, that's not particularly useful because that's still too dangerous or inconsistent experience. I think it's really going from like 90 to 95 to 99 to a hundred percent is where a lot of the work is. And in certain areas where we've bitten off experiences or want to use AI, it's unclear whether you can ultimately get there, which is why At B-Twelve, we're going after kind of what we call human-assisted AI, where we always think people will be in the loop. And as there's more technology, more capabilities, more compute power, we'll figure out ways in which automation can play a role. But it doesn't necessarily require us to a hundred percent automate the complex workflows that we go after. And I think each startup is different. In a lot of cases, there are people behind the machine that are masked or not put at the forefront. And in In certain cases, like, those people aren't going anywhere anytime soon if you want to create the type of experience I think consumers want.

AI assessment note: “In a lot of cases, there are people behind the machine that are masked”

Q In terms of algorithm training there, how do you approach that with B-Twelve?

A So I think the way that we often think about things is before you start out and build out really fancy technology and tooling and algorithms, We want to really make sure we internalize workflows ourselves. So with our first product, the first thing we did is just really understand how exactly the website and web design process worked. We did things on our own, and just were able to create very simple rules around how we wanted things to work from a machine output perspective. And then as we start to get more and more scale, we're able to use a network of experts that we have to think about ways to get very specialized data sets specific to web design. And so a lot of it is, I think first is understanding the end output that you're trying to get to, and then thinking about how do you collect data from actual human and person usage that can make the experience better.

AI assessment note: “ways to get very specialized data sets specific to web design”

Q Now, I'm super intrigued. As a young person in venture at the time, did you find it quite a lonely place? Often for the juniors in venture, it's often spent DDing, researching, and spending a lot of time on your own. Did you find this to be challenging or a lonely place?

A Not really. I think at GC, there's a really great cohort of associates and principals, and so we had a really great group of people to work with. It is a little bit more of a solo sport, because anytime you're doing a deal, you're the person at the firm that's primarily responsible for it. I think compared to when I started in 2010 to when I left in 2015, there's just been an explosion of young people in the venture industry, and people are doing really incredible things. So it's very different than a startup Or a bigger company where, you know, day in and day out work with a team. But there's luckily a great support network, and the folks at GC have been, you know, very close mentors, friends, and partners through the whole process. So I think there was a lot of people to kind of communicate with. And then above and beyond that, you're spending so much time meeting with entrepreneurs, supporting other folks in the community. So you end up spending a lot of time meeting with people and interacting with people. So it was, it was always something that kind of like made the, the Process, and even some of the more analytic stuff, analytics-driven work, feel human, because there was a lot of personal interaction.

AI assessment note: “Not really. I think at GC, there's a really great cohort”

Q So would you say the incumbency advantage with regards to data set is offset through the further Specialization of data sets, as you said there, into actually how users behave on websites. I don't know, heat maps as to where they click potentially. Do you know what I mean? Is it the further niche that we go is where the startups will offset the incumbency data advantage?

A I think it really depends on the particular space or application, but it certainly is true that a lot of the data that bigger companies have captured or will capture may or may not be the types of things that actually create the product experiences that startups Build or technology companies build and that consumers ultimately love. So I think there is an ability to do interesting things around creating proprietary and unique data sets that are way more applicable to the types of products that we want to build. In a lot of cases, collecting that data can be quite messy. You know, you use actual people to train an algorithm or use actual people to collect very unique and specific data. And whether you're a big company or a small company, that can be challenging. And often as a small company, it's a little bit easier because you don't have The reputation risk and or the challenges that a bigger company might have around organizing a cohort of people that in their world is very small, but in a startup world could be, you know, a very important group of people to take care of and treat well.

AI assessment note: “creating proprietary and unique data sets that are way more applicable”

Q machine there, are you bullish on individual teams creating kind of proprietary and, and their own different algorithms, or whether we'll see your IBMs creating machine learning as a service, and that will really become a real offering? So what do you think? Do you think we'll see this kind of monopoly of machine learning as a service, or do you think we'll see individual teams Tailoring their own algorithms.

A Yes, I think it's certainly true that the big companies have aggregated massive amounts of machine learning and machine learning research capabilities, and one of the really nice things is that a lot of these companies are open sourcing and democratizing these services, so things like TensorFlow from Google, and the result is, you know, we, when we recruit engineers, I mean, even an out-of-school engineer can do something now that only specialists could have done a few years ago, so I feel like There is definitely a democratization happening around some of the specialized machine learning services, which I think is a really great thing for the ecosystem because it enables startups and companies to create real applications. There probably are some startups that will do breakthrough machine learning, research, and intelligence, but I see a lot of them getting scooped up by big companies right now, and it's, I think, very hard to compete against all the resource and horsepower that these big companies are putting into research, and the fact that they're willing to democratize and open source these types of tools. One of the areas where I feel like startups Can have a distinct advantage is creating unique data sets that do interesting things around on top of some of these machine learning algorithms. And so, for example, with our service at B-Twelve, one of the first applications w…

AI assessment note: “startups Can have a distinct advantage is creating unique data sets”

Q And then I want to finish today on the next five years for you and for B-Twelve. What's in the roadmap?

A Our kind of belief is that startups kind of earn their ticket to doing great things by building a product that someone loves. And to build a product that someone loves, you start narrow and really define who you're going after and what exactly you're doing. And so all eyes for us are focused on how do we build the best way in the world For individuals to build, manage, and optimize their website, something that's, you know, substantially better than what's out there in the market. As we continue to kind of make more and more progress in doing that, we'll start to use orchestra as an underlying technology to create additional services for, for businesses. And my dream is that five years from now, you'd have hundreds of thousands or millions of people coming to B-Twelve, finding really fulfilling and encouraging work, and we're enabling businesses to get access to the types of services that they could only dream of today.

AI assessment note: “my dream is that five years from now, you'd have hundreds of thousands”

Q I guess for me then, what services would we like to see democratized that are currently for the one percent? And also, we mentioned the models of work there. What does this mean in terms of changing it and making new models of work?

A Yeah, so in our case, we're focused on B to B products and services, and we focus after, we go after the small to mid-sized business segment. So if you're a mid-sized business, there's a lot of things that an average mid-sized or small business owner probably has a checklist of dozens of things that they would love to get done, and they either don't have the personnel to do it, It can be cost prohibitive or it's just too complicated. And by having kind of an easier way to get access to these services or products, we think ultimately they'll be able to do things that might have felt out of reach in the past. And an example of that would be our product with websites. So the average mid-sized business probably doesn't have an online presence or website that's really doing them a lot of favors online. You know, they're, they, most businesses don't feel like they're necessarily winning online with their digital presence. Part of it is it can be really hard to find good expertise around the space. It's really time intensive to manage on your own. There's a lot of specialized skill set from design to digital marketing that you have to understand. With a service like B-II, you get access to that expertise in a semi-automated way, and so you can have a real digital presence that allows your, enables your business to thrive online in a way that it might have not been able to have done in…

AI assessment note: “In terms of work models... human-assisted AI model is that you can take away”