Tech Brew Ride Home - (IHP) Microsoft CTO Kevin Scott

Episode Date: June 10, 2023

Kevin Scott is the current Chief Technology Officer of Microsoft. We talk about his entire career, how being an academic seemed to be his path before he transformed the ads system at Google. Then he revolutionized the entire advertising industry at AdMob; is credited by some people by saving LinkedIn from technical rot; and now, today, oversees Microsoft's efforts in AI, VR/AR all the future things. Fantastic conversation. Kevin's podcast is: Behind the Tech Originally Aired: May 2019 Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to the Internet History Podcast. I'm your host, as always, Brian McCullough. Kevin Scott is the chief technology officer at Microsoft.

Before that, he was at Google, where his work there was influential enough to earn him a Google Founders Award. And he was at AdMob, and he was at LinkedIn. And I believe, actually, he's the first person from LinkedIn that we've ever spoken to on the show. Today, we're going to talk about all of that. But, of course, CTO of Microsoft. We're going to get into AI and future tech and all that good stuff that Microsoft is at the forefront of these days as well. CTO of Microsoft currently, I'm pretty sure that's officially the highest ranking current executive at any tech company we've ever spoken to on this show. My thanks to all of the comms people at Microsoft who worked for many months to make this great episode happen.

Please enjoy this conversation with Microsoft's CTO Kevin Scott. Scott, thanks for coming on the Internet History Podcast. Thank you so much for inviting me. You grew up in rural Virginia, I believe, tobacco country? I did. What was the name of the town? I was just reading this right before we got on the air. Yeah, it's this little town called Gladys, Virginia.

It's in Camel County. And I think Gladys in 1972, when I was born. was population 250, and it's still about the same size in 2019. My family still lives there. Yeah, I was going to write down the name because that's such a pretty name for it. I was like, wow, that sounds awesome. But, all right, so normally the way I start out with people is, you know, their first computer or their first experience with computing.

And then depending on the generation, that can range from, oh, Commodore 64 to, oh yes, the first time I submitted punch cards. So how about you? How did you, like literally first experience with a computer that you can remember? Yeah, I think I'm in the generation of folks where probably my first program that I ever wrote was the 10 print. Kevin is awesome. 20 go to 10 on, you know, some sort of it was probably like a Commodore, Vic 20 or something like that that was in a department store attached to a little cathode ray

like 13 inch TV so like this was in the early 80s when you know sort of before the the IBM Microsoft flavor personal computing but you know sort of the wave of innovation that happened after Apple had invented the you know the Apple

and it's like I was immediately captivated because it was the other thing that was happening at the time is console arcade games had started to show up everywhere like they were in the lobbies of department stores

and in convenience stores and you have video arcades and I just loved these games and I love these little consumer great personal computers mostly because you could play video games on them and my interest

in programming was like, man, I really want to be able to figure out how to make my own games. And like I've been a tinkerer ever since I was a tiny, tiny little child. Like my mother struggled with me because I would take everything in the house apart to try to figure out how it worked. And when I saw a computer for the first time, I was like, this is the greatest thing that I've ever seen worthy of being taken apart and understood. So do you become a programmer as a child? Yeah, I did. I around

I think it was probably 12 or 13 years old when I wrote my first real programs and it wasn't like they were works of art. So I wrote little video games. I wrote at one point when I was a teenager. I was really into Dungeons and Dragons

and I was always a dungeon master. So I encoded all of the rules of the Dungeon Master's guide into this little base, actually one, it was sort of a big, for me, basic program, so that I could sit there very awkward by

today's standards with my Radio Shack Color Computer 2, tethered to a, you know, a bulky 13-inch TV, you know, sitting on a coffee table, totally disrupting the ambiance of the Dungeons and Dragons game with... Everybody around the table.

You ran your games using the computer. I did. Wow, that's cool. Yeah. Yeah, go ahead. And I think, like, the first commercially useful piece of code that I wrote is my dad, it was a small-scale entrepreneur. He owned a few small construction companies in Central Virginia when I was

little and I wrote all the payroll software for him. So I like the thing that kept track of people's time and printed their checks out every week. It's surprising to me how common that that story is for people of our generation where you get your start by like helping your your mom. She's a teacher. So help her do her grades or your dad. Yeah. It's surprising how common that story is. So, because we're roughly, like I'm saying, about the same age. So I would imagine that your first online experiences might have been BBSs or maybe the early online services like Prodigy or CompuServe. Tell me about your first online stuff. Yeah, I played around a little bit with BBS.

It was interesting for us because Gladys is very rural and so almost everywhere interesting. was a long-distance telephone call, and like, we were fairly, you know, on the poor end of the spectrum most of my childhood. So I didn't have a lot of money to spend on telephone bills. And, like, I wasn't into the, you know, phone freaking stuff, like trying to illegally get yourself access to long-distance circuit so you could call into BBSs. that were too expensive to get out otherwise. So there were some stuff that I did locally, but I think the first time that I had

like real connectivity to the broader world of people messing around with computers was with CompuServe. CompiServe, of course, being the, it was known as almost the more nerdy of the services Like Prodigy was sort of trying to be general consumer, but CompuServe was where you were if you were into computing and nerdy stuff. Yeah. I mean, it had the stuff that was on CompuServe was the set of things that interested me. Like, you know, by the time I was a teenager, like the magazines that I would buy at the bookstore were Dr. Dobbs Journal.

And like I buy the nerdier of the computing magazines. Like I love Bight magazine. was absolutely great because it was, you know, in my mind, like the more technical of all of the many, many computing magazines that were being published at the time. But like Dr. Dobbs Journal was like my favorite thing in the world. Like anything that could teach me new programming tricks, I was all about. And so that was most of my motivation for interest in being online at all is to connect with other people who were programmers and to sort of learn new tricks because I was always try to figure out, like, you'd sort of see something in a game or an application that

was sort of cool. And, like, I just immediately wanted to figure out how it was done. So was there never any question you were going to go to school for computer science? Oh, there were tons of questions. So neither of my parents went to college. And most of my high school experience, I just, or like middle school and grade school, like, I just sort of hated it at all. It was, I used to tell people if, like, hell is an actual thing for me, it would be, you know, sort of sitting in a seventh grade classroom for eternity. It was just so boring. And, you know, I got lucky on a couple of dimensions. You know, my dad, my dad, neither my mom nor my dad went to college. My mom after she graduated high school, moved from this little place in Campbell County, Virginia, to live with my aunt in Richmond, which is the capital of Virginia for a little while. And she went to what at the time they called a secretarial school, which is sort of like a community college.

And she, you know, I think went there for a year or two and then became a bank teller. And my dad had been accepted to go to college. Actually, the same college I wound up going to Lynchburg College, which renamed itself to Lynchburg University. But he decided not to go because he hadn't enjoyed school very much either. And like immediately after he declined enrollment, he got drafted and went to Vietnam. And, you know, I think he regretted most of his life. not going to school and, you know, he just sort of walked around with a lot of, a lot of expressed regret about the set of turns his life had made. And, you know, he put an enormous amount of pressure on me to go to college. And he didn't care for what. But, like, he just sort of knew that, like, I was going to go to college and not make the same last minute mistake that he did.

And so, you know, I would work for him at his construction companies when I was a teenager just to have gas money. And he would give me the crappiest jobs. And he would tell me he was giving me the crappiest jobs. It's like, okay, like you go run the jackhammer to break up this concrete floor in this dang hot basement of this construction project we're running. or like you get to carry the shingles up the ladder onto a hot roof in the middle of August or you get to go push the wheelbarrows full of mixed mortar up the hill to the brick layers. And he was like, I'm going to teach you how miserable, you know, this can be. And like, I actually don't think that it's miserable work.

Like I do a lot of stuff with my hands right now. And like, I think there's an incredible amount of satisfaction and dignity and all that stuff. But, like, this was my dad sort of projecting, projecting onto me. And so, like, I just knew that I was going to college. And I had a lot of interests, and I still do. Like, anybody who knows me, like, you know, this sort of looks at my bucket of interest. And they're like, oh, this is sort of crazy. Like, you're all over the place.

And so for me, like, I had been really, interested in illustration and design when I was a teenager. And so I thought for a while, like, maybe I'm going to go to school to become a graphic designer. And when I actually got to college, like, I had this tug of war the whole time that I was there between English literature and computer science. Like, I was equally interested in both. So it turns out it was better at one than the other.

But I remember getting to the end of my bachelor's degree. And the discussion I was having with everyone is like, oh, you should go get a PhD in English literature. And, you know, like this other camp of folks, like, oh, you should go get a PhD in computer science. And I wound up actually not going immediately to graduate school because I was basically tired of being as poor as I was. and I went out and got a job and tried to save a little bit of money before I went to grad school. What was that job? So this was, you know, one of a great many happy accidents in my life. So a couple of pivotal things happened to me in my last year high school.

So I got chosen to go to the Central Virginia Governor's School. for science and technology, which was this magnet school that the state ran in central Virginia, and they would take two students from each high school in basically this multi-county area. Campbell County was part of it, and so there were, I can't remember exactly what the number was, but it seemed to be like there were sort of 40 kids or so in each of 11th and 12th grade made up from the top two science and tech students from all of these high schools. And so my friend and I got nominated to go when we were both in 11th grade, and they said they would only take one of us.

And my friend wound up. going that first year and then the second year they decided that they would take two of us. And so, like, I got to go to the Science and Technology High School. And it was like this really amazing experience. And the guy who taught me computer science there went to church with this guy who, starting in the summer that I graduated high school was starting his first company. Like he was an electrical engineer and like he had designed power systems and like electronic control systems forever. Like he was starting his company to do some consulting on that sort of work,

but to be a contract manufacturer to make printed circuit board assemblies for folks who needed small runs of things. You know, like imagine a Hamilton Beach who needs 10,000 circuit boards for blenders or like the company that, you know, needs like a thousand circuit boards for a batch of laundromat commercial dryers that they make. And so because my computer science teacher at this high school had known this budding entrepreneur man named Robert Roberts, Robert was looking for some summer help. So it was him and two people that he had worked with who, you know, were sort of the, the starting founding team at this company. And like I was I was flunky number one. And so I went in there and like I just did everything.

And it was the thing like I worked that summer. And then they were like, oh yeah, like you can continue working part time while you're in school. And so I work part time there probably 20 hours a week the entire time I was in college. And I worked full time in the summers just to try to make enough money to. help pay for school because my family didn't have enough money to pay for my college. And then when I graduated, I just sort of decided to work there full time for a couple of years. And it was just this brilliant thing. I learned so much about entrepreneurship and running a business and flexibility and grit. And then the really great thing is, like, the really great thing is, like,

Robert is one of the best engineers that I have ever met, like wickedly, wickedly smart and like a really great teacher, which is like this fantastic gift to have in your career, like just sort of being able to convey very complicated concepts to other folks. And so I effectively, even though it was like getting a degree in computer science, I learned to be an electronics engineer. year at the same time working for this company called Electronic Design and Manufacturing. Well, and then for many years, like you mentioned, you go on to get your MS at Wake Forest, but for many years also, unless I'm wrong, you're also in essentially academic computer science, like you're a research assistant and things like that. What are you working on on the academic side. Yeah, so I wanted my goal, like coming out of, coming out of like that high school experience I had at the Central Virginia governor's school, like I had decided I wanted to be a

computer science. And like not only had I decided to be a computer scientist, like I was really captivated with compilers and programming languages. So like I decided I wanted to be like a compiler programming language specialist. And I don't know like how God's earth like an 18 year old figures that out. But that was, that was sort of the way my brain worked. And, you know,

I took a couple of years off between bachelor's, like get my bachelor's degree and then going to grad school. But yeah, after Wake Forest, I started the PhD program at University of Virginia

and passed my qualifying exams and started working on a, on a dissertation. And most of the work that I was doing there was on something that sounds fairly arcane. It's this stuff that we and a bunch of other people, you know, had a lot of, a lot of enthusiasm and energy around in the late 90s and early 2000s called Dynamic Translation. And so the basic idea is if you take a binary program that's running on a micropower,

processor. You know, what the microprocessor is doing is it sort of like takes the binary program instruction by instruction and sort of runs it, you know, through the, you know, the silicon on the microprocessor to, you know, execute your program. And the idea with dynamic translation is you can insert another piece of software between program and the microprocessor. or to translate those instructions on the fly to something else to accomplish some other purpose. So, like, you're basically rewriting a binary program on the fly.

And my work was on how you can very efficiently write a framework that would let you build dynamic translation applications. So, like, you can use it for a bunch of stuff, like security. You can use it to try to do dynamic optimization to sort of. examine a program's behavior as it's running and to sort of optimize it to perform better based on those observations. And so there's just this wide range of applications that you can potentially write, but they're very hard programs to write in general just because you're sort of mucking around in the deepest bowels of the machine, you know, at the intersection of

the operating system and, you know, the very back-ins. output of compilers and like, you know, the raw instruction set architecture and, you know, like the computer architecture of these microprocessors. So the framework that I wrote was turned out to actually be somewhat useful. Like there are a bunch of folks who ultimately went on to use the tools that I and a handful of other folks. folks built to actually do research and some practical things using this approach to software. Correct me if I'm wrong, but somewhere in there, weren't you also briefly a research intern at Microsoft? I was. Yeah, my PhD advisor, an awesome, awesome guy named Jack Davidson, who is like one of the

real titans of compilers and programming languages and sort of low-level system software. So he's fellowed ACM. He's like a whole bunch of the code that he wrote when he was a PhD student influenced the original design

of the Gnou C compiler, GCC. And And, yeah, Jack went on sabbatical to Microsoft Research, and he took me with him. And, like, that was a super fun experience. That was 2001, I think. And I take you with him, like, to Seattle? Yeah, so I went. He was at Microsoft Research in Redmond, and he, like, the way academia works is, like, every seven years or so, you get to,

take time off to go do something, you know, different like you might teach in another school, you might go to a corporate research lab, you might do something else entirely, you know, with the purpose of getting yourself reinvigorated and exposed to new and interesting things. So he moved his whole family to Redmond to be an MSR to work in this little group in MSR at the time called the Programming Language Systems Group. and he got me a summer internship there at MSR, for which I am eternally grateful. And I packed all of my crap into the back of my Volkswagen GTI,

and I drove from Charlottesville, Virginia to Redmond, Washington, and spent several months there working in that research group. Well, how do you get to Google then? Why abandon this academic research track? Well, at some point in all of this, I've met my wife. So she was a PhD student in the history department at the University of Virginia when we met. And we met right after she took her qualifying exams. And she's our early modern European historian.

Like her specialty is 16th and 17th century, German history. And as soon as she passed her qualifying exams, like you start working on your dissertation. And for a historian, working on your dissertation means going to find your documents that are going to be the foundation of the dissertation that you write. And all of her documents were in Germany and, you know, sort of the German-speaking places in Europe. And so we met. She just passed her qualifying exam.

She already had a date by which she was going to pack all of her stuff up and moved to Germany. And, like, we fell in love and, like, had this quandary of. of like what on God's earth are we going to do now? And so I, through what still is like, seemed like another one of these extraordinary pieces of luck that I benefited from, I reached out to the, so she was going to this town in Germany called Gottingen, and then I reached out to the computer science department

at the University of Gottingen and said, like, hey, I'm coming to Germany with my girlfriend. I don't really have anything to do when I'm here other than, you know, sort of spent a bunch time sending emails back home to my advisor about my dissertation research. Like, is there anything to do in your department? And he was like, oh, this is great. Like, we actually just started this department. And, like, we're looking for lecturers.

You're hired, which evidently never happens in German academia. And so, like, I had this job lecturing computer science at this new CS department, the University of Götting. And, like, it was sort of, like, I didn't even, like, fully appreciate it was sort of strange that, you know, this is 2002, I guess. So it's sort of odd to me that, like, the, you know, the University of Götting and didn't have a computer science department at this point because University of Götting is, like, one of the best most stories. academic institutions in the entire world. Gauss was an astronomy professor there. Like Heisenberg was there. It was this incredible institution.

I was lecturing computer science at the University of Göttingen. And my wife and I wound up staying in Germany for about a year and a half. And, you know, we had, before I even went with my wife to, my girlfriend at the time, to Germany, like, I had started to get a little bit skeptical about the impact that I could have in academia as a computer scientist. And I wasn't really happy with the way the incentive structures were working there. You know, you spent a lot of time writing, writing a lot of research papers. Like there's this definite incentive to be a little bit incremental. Like the safest thing to do is to like try to push something by 5%. It's like, you know, very dangerous if you're trying, your PhD student trying to get a tenure track job or you're a, you know, you are in a tenure track job, actually trying to get tenure.

There's a formula you can follow where you go get really good at writing grant applications and pulling in money and getting a research program where you're sort of going to understand whether or not you will be able to get a reasonable set of outcomes. And that's not to say that there aren't people in academia that take enormous risk, but the risk feels sort of enormous because. you know, if you spend, you know, as an assistant professor, five years, your first five years is an assistant professor, like, trying to do something incredibly bold that takes all of your resources and all of your attention, and it doesn't work. Like, you're in real trouble in terms of, you know, getting tenure. And so, you know, I was looking at this and, like, I published a bunch of research papers and, you know, the, The interesting part of my research even seemed like the platform pieces, you know, like the fact that I built something that made it easier for like a bunch of other people to build interesting things. And like that was a harder paper to get published than like writing about like, okay, well, like I use these tools to like make this set of synthetic benchmarks, like some number of basis point. better than the previous optimization tools. And so there's all of that, and I really loves teaching. Like, that was the thing that sort of invigorated me.

You know, I got even got a until PhD fellowship at one point that basically gave me the, like, the funding where I didn't need to teach. and instead, like, I decided to, I got my PhD advisor to let me, like, run a graduate seminar for the other grad students. And, like, everybody thought I was a nut. They're like, you know, you shouldn't, you just got this amazing gift where you don't have to do this annoying teaching business. You know, you should, you should be, like, writing those last few chapters or your dissertation. and like that just it was like a bag of stuff that was getting less and less interesting to me over time. And then I was in Germany with my wife and like she went through a real similar transformation. The funny thing with her is I'm a super, super introvert.

Like it's, it's, and I still am. It's really for me to, you know, like being crowds of people and to chit chat. And, like, you know, I will, like, get to some point with these, you know, sort of browny emotion, social interactions. And then, like, I'll just have to go be by myself somewhere to recharge. My wife is exactly opposite. She's, like, a mega, mega extroverts. So, like, she will, she has to be around lots of people and having lots of these interactions in order for her to, like, have energy. and the longer she's alone and by herself,

like the less well she is. And so it was like really odd. Like, you know, ironically, like you sort of look outside versus in. You say, oh, like historians, like humanities, like these people are like all, you know, extroverts. And like that's an extroverted thing to do. And like computer science and software engineering, oh, that's, you know, just sitting behind a screen. And like turned out that like computer science is, you know, at least in our narrow experience, my flavor of computer science was way more collaborative

than her flavor of history. So she got to Germany and she was like sitting all day long in these archives in complete silence surrounded by these, you know, musty documents with no human interaction. And she'd come home at the end of the day from this and just be like, ugh, like miserable. And so, and she also, really enjoyed teaching. And like we both knew that like if we stayed in academia and like, you know, the goal was the tenure track job at the like big important research university that like we were going to be pushing ourselves in directions that we did enjoy.

And so we got to the end of our funding, the grants that were supporting us being in Germany. and like we had this decision to make. Okay, we can go apply for more stuff and like double down. And, you know, she still needed to be in Germany to finish her research so she could write her dissertation. And, you know, she went and applied for a bunch of these things. And I went and looked for jobs in Germany that, you know, you could sort of get without having completed your PhD. And, you know, being an expat. And like we weren't in, you know, a big place like Berlin or Munich or Hamburg.

We were, you know, like she was thinking about, you know, going to a place. And like, you know, I was looking at the like Fronhofer Institute and Fraunhofe Institute in Kaiserslauton. And, you know, like we did a bunch of that stuff. And then I was like, okay, well, like we should also like look at some alternatives. And I sent my, I sent my resume. I kid you not to Google. And I didn't even know why I was sending my resume there. I knew that a bunch of the compiler people that I really respected, like Orsa Hotsla,

and, you know, who had been a PhD student of one of my PhD advisor's colleagues. And, you know, there were folks like Jeff Dean. whose PhD work was in compiler optimization, like Alan Eustace, who eventually was the head of engineering at Google was there. And I'd actually cited his work in my incomplete dissertation and in a bunch of the papers that I wrote. And I just sent my resume and I was like, all right, well, if these interesting people are working there,

like maybe there's something interesting for someone like me to do. And I got a call back from them and they wanted to interview me. And I flew out for the interviews and they were great. They very thoughtfully, this is in 2003, so like year before the IPO. And they had very thoughtfully looked at my background and they stacked the entire interview slate with compiler and programming language people. And I was like having these great conversations. I mean, they were interviewing me and like asking me all this tough stuff, but like all of

it seemed super fun. And I like walked away from that. And I was like, yeah, this is the this is the thing I ought to go do. Like I still didn't understand like how Google was going to make any money or like whether this was a good decision or like, you know, what what big contribution I was going to be able to make to either search or advertising. But I knew that it was just a bunch of really wickedly smart people there. And I knew that I enjoyed interacting with them.

And I was like, all right, I've got good this to go. And then on top of that, I got the chance to be one of the first engineers. Like, I think I was, like, number 10 or number 11 in the New York City office. And so, like, that really sealed the deal. It's like, okay, like, my wife and I, like, really, like, at the time, enjoyed these urban settings. It's like, we, you know, we lived in Germany. like we didn't have a car, like we could walk or bike everywhere. So like being in Manhattan, like, sounded just great.

And so that's what we did. Well, and then you do, it's not the search side, it's the ad side that you work on, right? And you're saying this is a year before the IPO, but also this is maybe a year or two into Google, even having an advertising platform. Yeah, no, it had an advertising platform for a while. Well, right, but beyond the Tim Armstrong stuff, I'm saying, like, this is early on in the AdWords, AdSense rolling out, period. This is before AdSense had rolled out, and like when AdWords was, like, in its early days of taking off. Like, I was part of a whole bunch of people that Google was just starting to hire in 2003 because that business was starting to start. to scale up and like they had both the you know the funding and the need to bring in a whole bunch of

people and so like I actually you know I worked a bunch in both search and ads but like my first my first real project that I did at Google was was an ads one and then you go on to lead the the ad quality engineering team yeah I let a big chunk of of the, so like ads quality was the team at Google that built all of the stuff that effectively like ranks and filters ads. So it's like the CTR predictions stuff and, you know, which you actually have to have for a like a second price auction to work. You know, when you're, when you, when you selected a set of candidate ads, like you actually have to like predict. what the likelihood that someone is going to click on them in order to, and you multiply that by the bid price to basically, along with the other participants in the auction to figure out what

the ad ranking is. So, like, you've got this big system for doing CTR prediction, which is like economic necessity, and then like a whole bunch of other things. Like, you know, when when should an ad not show up in the auction because like its quality is too low or um you know when are people trying to play like uh low quality arbitrage games uh in the system so it's like a whole bunch of stuff that this team did and like i ran uh i ran a big chunk of it i didn't run the um i didn't run the team of economists and statisticians that were also technically part of the team i work for a guy named Mike Framkin, who actually ran the whole ads quality team. And Mike is awesome.

He's like one of the best bosses I've ever had. Just an incredible guy. Ad mob. So I think you're at Google the first time for a little maybe under four years even. Three years and nine months, I believe. Okay, okay. So what interested you in ad mob and why I head over there? You know, I tend it at the time, and I still do, like I spent a lot of the time to think about what it is that I want to accomplish over very long periods of time.

And then I try to make a plan that helps me get there. And so I had spent basically from age 17 through age 30, 31, and like my entire goal in life was to be a professor. And when that changed, I needed to get a new plan. And so I was at Google for about a year, and I figured out that maybe the thing that I can be good at that lines up with my interest is not just being an engineer, but helping other engineers figure out how to do their best work and how they can make an impact in the world and how they can make an impact in the world and how they can. achieve their own goals and career ambitions. So, like, you know, it started because I was in, I was at Google. And we, you know, we just had all of these brilliant people there. And, you know, we were hiring folks so fast that we didn't always have this prescriptive thing to tell them to go do.

And like we gave people a bunch of freedom about what they chose to work on. And so I thought, okay, like, I might be, it might be worthwhile for me to help people point themselves at things that are going to, like, be at the intersection of, like, what's going to produce value for Google and, like, what's going to align with their interests and expertise. Like, that's sort of the, you know, the sweet spot, you know, so to speak. And, like, I've been, you know, lucky that, you know, the thing that motivated me to go into industry in the first place was, like, I wanted to do something where I could, scratch a whole bunch of technical itches, but like in a way where like it was going to have measurable impact. And so like I decided at that point that I was going to become a manager. And as soon as I decided that I was going to lead teams of engineers, like what I wanted to do was like, look, you know, my new goal is to be the head of engineering at like a big internet company. And I knew that there was no way that I was ever going to be the head of engineering at Google. And so to

me, like I was at Google and like I tried to expose myself to like everything humanly possible to like help me have skills to, you know, and pattern matching and coping mechanisms for being the head of an engineering group, like the person with whom the, you know, the buck stops and, you know, who like has the, you know, the whole account. accountability for, you know, for an engineering team. And like, I got comfortable at some point. And I was like, okay, well, now, like, I want a head of engineering job. And I was like, okay, I think I can probably go get one at a startup. And I looked around this in 2007 at like all of the interesting things that were going on. And I was like, oh, this is like, this is an interesting thing. Like I'm sort of jumping into the frying pan. Like there's a whole bunch of. of stuff about being a head of engineering at a startup and about startups themselves that I probably that I know that I don't understand, but like at least I understand advertising. At least I understand, you know, how to lead an engineering team this size because like I led bigger teams at Google and like when I got there, it was like about 10, 10 people in the engineering team. Like I knew from just a leadership perspective, I could handle that. And, and And, you know, like, I could, basically that would give me enough of a buffer where I could sort of learn quickly all of the things that I didn't know how to do.

And I knew also just because, like, I'd been sort of seeing some of the momentum in the mobile ecosystem. I knew that this was going to be an important thing. And, like, the problem that we were trying to solve at Ab mob, like, really, really spoke to me. wasn't that like, okay, let's go put some little irritating rectangular boxes on a small screen. It was, you know, if you'll sort of remember back, like I joined, I joined AdMob before the iPhone had really launched, like it had been announced, but like not, not launched. And like, our entire business at that point were text ads in WML applications and like almost all of the, you know, running up to this, the way that a mobile application got distribution is

like they had to cut a deal with a carrier to appear on their deck. So like this is sort of like the home screen for the phone. There's no app stores. Those things don't exist yet. No, none of that. And you had, it was sort of like this major impediment to innovation happening because, you know, there were all of these chicken and egg problems. Like, you know, Carrier is not going to talk to you unless you've got, like, some credible business story. And, like, a whole lot of the great ideas that have happened in mobile over, you know, the intervening years, you know, start off.

And, like, you can't really prove that they're going to be the success than they are. So it was just like this, like, innovation really couldn't happen as fast. it needed to. And so Amob was all about like how can we give application developers a distribution mechanism where they can find an audience for their apps and services that they're building. And once they have that audience, like what's a mechanism we can give them to monetize it? And so it was basically about us trying to figure out how to deliver those two essential ecosystem components so that innovation could happen. And like the thing that felt really great about ABMob is that, you know, we had this point, you know, a few years after I started where, you know, like it was just the crazy fastest growing thing I've been part of.

Like the business was doubling at this ridiculous doubling interval. And like a whole bunch of the, you know, the mobile companies that were getting funded at that time, like their, you know, their monetization part of their. business plan and their pitch decks was like ad mob and like that was great like we were sort of doing exactly exactly what we wanted to do and it's awesome like the team that we had there it's uh you know the people what the people have gone on to do is sort of awesome so my uh my uh my uh director of engineering there is now the ctto at instacart uh Cheryl Dowell Rompul who was uh our CFO was a CFO at And she's the CFO at Confluent right now, which is this ridiculously successful company that is sort of bringing the whole notion of streaming data systems and real-time analytics and processing to the whole world based on Kafka. Jason Spiro is like one of the big advertising executives at Google now.

Omar, the founder, is like a great investor at Sequoia. I mean, it's just, you know, and in and, you know, like Kamakshi Shrinivasa, and, like, one of, you know, one of the incredible data scientists, like, she had PhD from, you know, Stanford and double-E information theory. Like, she's founded this company called Drawbridge. Like, it's just, like, that team punched way above its weight. I'm like way, way prouder of all of them than, you know, like what we were able to do, do for the ecosystem than like a great many things that I've ever done in my life.

An ad mob mafia, as it were. Yeah. The PayPal mafia, yeah. I'm going to jump ahead to LinkedIn. You're the first person I think I've talked to from LinkedIn. So a really basic question, you know, you've worked at Google and, Microsoft and ad mob, so startups, all sorts of things. Just how is LinkedIn unique or like even from a cultural level?

Describe, describe to me LinkedIn? Yeah, LinkedIn is, I mean, look, I'm highly biased because, you know, I joined LinkedIn when, like, in 2011 before the IPO. And, like, my job was to, like, build that engineering team into, you know, like, a big thing. It basically was the job that I had been looking for for a very long while, like, head of engineering at, like, an important internet company. And so, like, I'm super biased about the culture because, like, you know, sort of the culture that, you know, my, the leaders and great engineers there, you know, in, like, this sort of full cooperating. with, you know, with Jeff and the rest of the company. Like, it's just sort of the place

that we wanted to be. And so, you know, we one of the, one of the things that we focused on a lot is you know, LinkedIn's business, like, never had the, like,

gigantic margins that some of the other big companies had. But, like, we had many of the same technology challenges. So we were sort of scaling at this incredible rate and huge amounts of data. And so, you know, like you take those two things together. Like we had this like crazy need for focus and prioritization. And so one of the, like we just sort of designed, you know,

designed a culture around us being able to get the, best possible technical and product outcomes that we could get from what always seemed to us like a constrained number of resources. So, like, collaboration, for instance, is, like, super important in an environment like that. You just cannot afford to have people working at odds with one another, duplicating a bunch of effort, you know, like trying to, you know, sort of advanced. their agendas through these sort of parallel and competing paths. You just really have to get people to come together as a team and to recognize that the most important thing is for the entire team, like not to think is like, you know, sort of a random confederation of slightly warring tribes. But, like, you know, you're one team, you're one team at 250 people, you're one team at 1,000 people, you're one team at 3,000 people.

And, like, you've got to always keep in mind the top level objective that you're trying to accomplish and, like, be willing to play your part in accomplishing that objective, even if it means locally suboptimizing what you're doing for some period of time. I saw, just specifically to this, I saw, you know, there's a Businessweek article. and a Business Insider article that essentially say to the level of, like, you institute a code freeze at some point that they say, in these articles, like, essentially saved LinkedIn, as they say, maybe that's hyperbolic. But, like, you had to come in and do things to such a level that, like, look, we got to get our eggs in order if this company's going to move forward. Yeah, look, we, I don't know, you know, who saved link, like, it's no one individual ever. saves anything. Like, that's just madness to think about. But the thing that we did know, like, we were in a precarious situation when I joined the company in 2011. Like, we had, you know, we, we had a bunch of different product engineering silos that weren't working well together.

And, like, we had a software development infrastructure that basically couldn't support our ambition. So we were, you know, it was just all kinds of craziness. Like we, you know, at some point we decided to go to a service-oriented architecture, which is like a great thing. Like, you know, service-oriented architecture is meaning that you've got things sort of, you know, factored into a set of units that have independent scaling characteristics. So, like, you can, you should be able to operate them independently of one another, like scale them independently of one another.

other and like they're, you know, like small enough where they, you know, have some sort of uniformity about them that let you, you know, reason about them, like, as an isolated thing, not necessarily in the context of the whole. But, like, we, you know, we'd sort of split this thing up into a thing where we had more services in production than we had actual engineers, like the services outnumbered the engineers. And they weren't independent of each other. Like, you couldn't deploy like one service independent of the rest and yeah so there was that like we were doing this like feature branch based uh development and like we were we would release the site like all of these services in like one big bang like once every two weeks yeah so when i started like i think it was

wednesday like every wednesday night like you would expect that you know 100 plus people were going to spend all night, like, getting the code that, you know, that they had slammed down onto this integration branch and spent the previous, you know, like multiple weeks trying to qualify. You know, you would then push this thing out into production. And like you'd have these handwritten deployment plans, like on wiki pages where it's like you do these 110 steps to deploy everything and they're invariably in each one of the deployment plans. There'd be some point where it's like, all right, this is the point of no returns. Like we can't even like fail all of this back if it doesn't work.

Like if we have a problem after this, like we're going to have to fix forward. And it's just, you know, I could go on for like a really long time sort of describing like all of this stuff. But like the net of it was that it was taking when I got there up to a month for someone to check a piece of code in to the time that it showed up on the site. And that period was increasing almost as a function of the number of engineers that we were hiring. And, you know, when I started there, like, the IPO was imminent. And so, like, we were just in this quandary. It's like, look, you know, we've got to keep growing the business. Like, we've got all of this ambition, like, all the things that we want to build.

It's getting harder and harder to build them, not easier and easier. And, like, everybody, all the engineers, like, you know, are super frustrated with how painful all of this is. They know that it doesn't need to be this hard, but like they couldn't figure a way to like get out of the trap. And, you know, we, we knew for sure that if we didn't have a credible plan for how we were going to fix it, that six months after the IPO, when the lockup period expired, every good person was going to leave in short order. And then, like, not only would we be busted, we would lack the capability to fix ourselves. And so, like, there was, like, a huge amount of urgency. You know, so, like, that co-free's, like, where I basically said, no new product development for, you know, for a couple of months at the end of 2011. Like, that sounded to, like, a lot of the, you know, the people at the company, like,

a really huge and risky thing. But, like, I was looking at the, you know, the real risk was. getting ourselves permanently into a state where, you know, like we couldn't build software. And so, like, to me, like, it wasn't that hard a decision to make. And, you know, like, we got things fixed. And, you know, this is also sort of a hard thing of going through a big transition like that. Like, if you have tried to fix a thing multiple times and failed, then, like, one of the big impediments to actually making a credible and successful attempt at fixing it is giving your team the confidence that

the next try isn't going to be the same failure that the last ones were. And so, like, we had to do a bunch of stuff to, like, help people feel like, yeah, we're really going to do this. Like, this is going to work this time. And that, you know, maybe out of all of it was the hardest thing because, like, technically, like, what we needed to do was sort of obvious. It was, like, like just getting everybody together and, you know, sort of pulling in the same direction. That was the, you know, sort of the challenging part in the end. And then, you know, like you sort of, you know, we did all of this in November and December. Like we, you know, like we literally at some point in, you know, in November, early December, like we had burned down the old, the bridges to the old way of doing things.

Like we couldn't build code anymore. and you know that was a that was a nerve-wracking moment but like we spun everything back up in January it wasn't it wasn't perfect in the beginning but the foundation was in place where we knew that if we continued to invest that like we actually could get it really good like really quickly and you know we were yeah more or less a completely different a completely different company in terms of engineer happiness and software development capability and product velocity by the middle of 2012. And if we hadn't done that, we would have been in real trouble, I think.

Well, then that brings us to present day because then the acquisition brings LinkedIn to Microsoft, brings you to Microsoft. And in my notes, I was going to ask you, you know, why take the CTO job at Microsoft when it's offered to you? But now, having spoken to you, I think it's pretty obvious. Like, this is the chance to create the culture, the engineering culture, the ability to empower engineers to do great work at, like, the biggest scale imaginable, right? Yes. So I'm glad that that seems, because that's exactly it.

It scratches this amazing itch that I have just in sort of terms of curiosity. I've never had a better higher bandwidth learning opportunity than the one that I've got right now just because we have so many brilliant people working on such a crazy breadth of things. There's so many impactful things happening across, like forget about Microsoft, but across the entire technology landscape right now. You know, like we're having, you know, just sort of moment after moment where, you know, like, it's not only as technology, like, having this, like, incredibly large impact on everyone's lives, I think people are sort of feeling the impact more and more in places that they didn't anticipate feeling it. And I think you have to, yeah, there's just a big responsibility in making sure that you are lining up all of those opportunities for tech to do interesting and great things with a real need to make sure that those investments are serving people well. the thing that I really loved about what Satya's done at the company and he's done a bunch of really sort of amazing things, but just simple things, like, reframing the mission of the company as empowering every individual and every organization on the planet to achieve more.

So, like, that sort of entirely lines up with my worldview. I think technology is a platform that should fuel other people's creativity, other people's ambition, other people's desire to make businesses to change the way that they're doing things and prove their lives. And I really do believe in this thing that Bill Gates said a whole bunch of years ago, where he defined a platform as like a thing that produces far. more economic value for the people building on top of the platform than it does for the people who built the platform. So I think there are things like AI, for instance, where, like, that's how

AI has to go. Like, AI has to be a platform for other people to build on top of that benefits them more than the, you know, than the companies who are building the AI infrastructure themselves. Like, just has to. Yeah, go ahead. Yeah, I was going to say, like, you know, to wrap it up, there's so many things that Microsoft is doing that we could talk about. But I let's do focus on AI, like, as a lens to talk about what Microsoft is doing right now. Like, I heard you say that there's almost this palpable excitement about the progress that's currently being made in the AI space right now. Like all of these years we've been told it's right around the corner, it's right around the corner.

And it almost feels like the corner is now and it's being turned. Yes, I think that's that's becoming increasingly I mean for some of it like I Yeah the stuff that I was doing Like years and years ago Like my very first project that at Google was a machine learning thing

It wasn't an incredibly sophisticated machine learning thing By today's standards But it was an ML thing And you know I've been doing this long enough now To just see from a developer's perspective how much easier it is today to build machine learning things than it was, how long ago was that, like 16 years ago now, 15, 16 years ago? I mean, just it's stunning, stunning.

Like, some of the things that, you know, were just, like, incredibly expensive time-consuming heavy lifts, like way back then. Like, you can imagine, you know, like a high school kid, you know, doing it a weekend at home. I mean, it's just, you know, just sort of crazy the amount of power that the AI platform is already putting it in the hands of folks. And, like, the, you know, there's, I think sort of, I mean, like, I shouldn't even be using the word AI because, like, it's almost a junk word, right? It's not like AI is one thing.

It is like a, you know, sort of an umbrella term for a very, very large number of, you know, sort of super specific things. But if you're thinking about machine learning, for instance, as this area of activity inside of the AI umbrella, you know, there's sort of two interesting buckets of things going on right now. So, like, I think the tools have gotten good enough where you've got a larger and growing population of practitioners have access to these tools and are. you know, sort of out there sort of filling in this long tail of, you know, the, the AI possibility curve. You know, so it's like these little applications that solve like really meaningful problems and like this incredible diversity of context. And, you know, the interesting thing there is like, how do you make that population, you know, expand, you know, where it was, you know, maybe hundreds or thousands of practitioners doing real, you know,

commercially impactful work 15 years ago and now maybe it's like, you know, large tens of thousands of people, but, you know, like, it's still like probably way more elite than it needs to be like, how can you get that to be hundreds of thousands or millions of developers who like have these machine learning techniques as part of their repertoire? And then like the exciting thing, like on that push is how do you? do you use the tools of machine learning themselves to allow people who actually don't have any programming experience at all to build things? So we bought this little company called Loeb a while back last year. And like the thing that Loeb did and is working on right now, they're sort of

pushing towards like having an open beta that every everybody can sign up for is like it's a simple enough machine learning development interface that you can use to build machine learning applications with no no programming expertise and like let me let me give you an example uh which i i give to everyone so one of the um one of the one of the founders of lobe uh mike mattis is a designer like very very good designer um you know he was worked Apple, worked Facebook, was like one of the first, like, I think maybe the first designer at Nest. It's just brilliant, brilliant designer. But, like, he's not a programmer by training. He lives in this off-the-grid house in Marin, and he's got a big water tank that buffers water for him. Like, he's solar powered, so, like, I think they use solar-generated electricity to run them. like a well pump.

But like you got to move the water into the tank so that at night you can use water because you don't have electricity to run the pump. And so he wanted to build a web app that would tell him how much water is in this tank at any point in time. Now like I'm an engineer

and the way that I would go solve this problem is I'd put a bunch of sensors in the tank. Like I'd figure out like what the mathematical relationship is between the sensor output and the water level. I write a whole bunch of code. Like I'd put a raspberry pie out there. I connected up to some complicated

internet crap. I write the web app. I just do all of this stuff. And it's a perfectly reasonable software or electrical engineering sort of way to solve the problem. But the way that Mike solved the problem with Loeb is I think way more approachable and intuitive. So he took a float and tied it to a rope through the rope over the end of the side of the tank and tied a weight to the other end of the rope. And so now he's got this thing where it's like a little more sophisticated than that.

There's probably a pulley in there. But like that's sort of the basic setup. And so like now you've got this thing where the weight as it moves up the side of the tank, it means that the float is lower and the water level is lower. And when the, you know, when the weight moves down on the side of the tank, it means that the float is rising and there's more water. And so he pointed a webcam at it and, you know, got a stream of images coming out of it. And then fed the webcam stuff into lobe and then like went in and made some manual measurements that corresponded to, a handful of those images and then press a button and lobes spits a model out that gives you a water level from the stream of images coming from the webcam.

And so, like, that is way more intuitively approachable than, like, all of the crap that I had to stuff in my head over the course of, like, many, many, many years in order to solve the problem, like, the engineering. engineering a sort of way. And so, you know, and Loeb is one among many sorts of tools where you're using the whole paradigm of machine learning and machine learning itself to sort of optimize some of the things in the back end, to be able to empower people to, you know, I don't even want to call it program, but like to develop software who like just haven't had any access to that superpower before. So that's one thing I think that's interesting about AI. And then the other thing that's really interesting

is almost like the other end of the spectrum is that there are just some remarkable breakthroughs that are happening right now on reinforcement and reinforcement learning and unsupervised learning where we're able to

basically remove some of the constraints that slow down the rate of progress for training big AI models. So like the constraint right now for like all practical purposes is like how how are you going to get enough label training data for supervisor, semi-supervised machine learning systems? And like how are you going to get enough compute to, like, train over large volumes of training data with, like, models that have very large numbers of parameters.

And so what folks are doing in some of these, in some, you know, reinforcement learning and unsupervised learning research and actual applications are removing that data constraint, that label data constraint. And so, you know, you think about some of these stunning results that folks have had in strategic gameplay, you know, like AlphaGo beating the, you know, the world's, you know, greatest go player or, you know, the OpenAI's Dota bot, you know, beating, you know, beating folks at Dota. and like even recently there have been like a bunch of breakthroughs in natural language understanding that are based on unsupervised learning like both of these buckets of things are leveraging the fact that you can build bigger and bigger and you know increasingly interesting models by just being able to apply more compute to the problem open a i's got this brilliant thing that they you know like this brilliant and in insightful is probably the better adjective blog post that they published last year, where they sort of show this progression of the amount of the number of petaflop days that have gone into training all of these remarkable AI models going all the way back to 2012.

And, you know, like they sort of plot this out. And, you know, like we basically each step in the progression of, you know, these all-inspe. inspiring results are coming along with like an order of magnitude increase in compute power. And so, you know, it's exciting because I think this, you know, increase in computing power is, like, going to continue to come for a while. And, you know, I think so I think both of those things are like maybe the two most interesting things happening in AI right now from just in terms of like what I think is going to produce interesting and useful things. for humans over the next, like let's call it, five to ten years. Well, a final question, because I think this is what I derailed you on that you were about to tell us about.

So final question, how should we think about Microsoft's efforts and ambitions in all of this stuff, you know, AI as the catch-all phrase in general, how should we think of what you guys are doing is different than what, say, other companies are doing in this space? Well, I think, you know, the thing that we're doing is, like, we started as a platform company, like, building software that, you know, enable more computing in the world. But more computing was not really the interesting thing. Like, what people did with the more computing was the interesting thing. So, you know, like we really do, like, through this, through this mission and through like all of the businesses that we're running right now, like we see ourselves as a platform. Like, we build things that help people to, you know, achieve the things that they're trying to accomplish in their lives.

And, like, our, you know, our business model aligns directly, directly to that. You know, like if you have an AI workload that, you know, that you want to. run, you know, some interesting AI application that you want to build. Like, you know, we have a cloud. Like, we've got a bunch of AI tools that we build. Like, you know, we're giving you the capability to go build some of those things. And, you know, it's like whether it's, you know, Azure ML or it's Azure IoT or whether it's, you know, Power BI for, you know, doing your analytics and visualizing your data or it's Microsoft Excel that you're using to build your models to understand how your business works or run your business or even if it's productivity

and collaboration. We build tools that help you get your stuff done. And I think like focusing on like just sort of empowering folks is really really sort of the exciting thing for me because Like, I think that's, you know, ultimately that's what technology has to do. Like, if you think about all of the, you know, the, all of the, you know, I don't want to sound overly grandiose, but, you know, like, the, you know, so technology and the progression of human society sort of go hand in hand, you know, like, we don't, we don't have the world that we have today. and like, you know, we can't support our, like, way of life without, like, this progression of technologies that we've had starting, you know, with fire and agriculture and, you know, the printing press and, you know, industrial manufacturing and electricity and refrigeration. And, you know, it's like you, sometimes I think we just sort of take a bunch of these technological things for granted. because we become so dependent on them and like they're so integral to our lives that they fade into the background and like you know we just sort of assume that you know they're always been there and they're always going to be there and you know i think they're you know like we sort of imagine you know our future um you know it's almost a certain thing that uh technological advancements that neither you nor i have fully conceived of or could even possible imagine are going to be the things that are

you know sort of receding into the you know the background and are sort of the default assumptions for you know like our kids and grandkids and you know subsequent generations and so you know like being a you know being a facilitator for some of that I think is you know like it's it's a huge huge responsibility. I think, you know, technologists in general, like, ought to be thinking about,

like, you know, how to, and it's not us. Like, we're not building the future. Like, we're building pieces

that other people should be using to build the future. So that's, that's sort of how I think about the world and, like, what we're doing at Microsoft. Hey, since this is a podcast,

should we mention your podcast real quick? Sure. So I, um, I've been for several months now recording a podcast called Behind the Tech. And it's actually sort of a similar idea to what we've been talking about all along. I get the opportunity to work with and talk with and collaborate with all of these interesting folks who are behind the scenes, like behind this technology that we're all increasingly used. using and, you know, dependent on and curious about and skeptical of, you know, like,

these are the folks who are building these things. And, like, I just find it inordinately interesting to hear their stories and to, you know, this is a podcast that, like, talks with some of these people. And, you know, it's what's always surprising to me is, and this is, like, one of the primary reasons that I started doing this podcast is there, you know, there's a certain archetype that I think we all have in our head of like what a, you know, what a technologist is or like what a, you know, what a Silicon Valley entrepreneur or a technology industry entrepreneur is. And like it's remarkably homogenous, like that archetype. And like when you actually go talk to the people, it's like they've got so many backgrounds and like they've come from so many different different, different,

places and different, you know, life histories, and they've had so many different journeys into their careers. And, you know, like, none of it's linear or predictable, you know, in a certain way. And, like, that to me is sort of, like, the fascinating things. Like, anybody can be a technologist. Like, anybody, I think, can do great things, like, using tech to, like, make interesting stuff. And, like, The podcast is about talking to those people about them. Well, Kevin, thank you for coming on this podcast and sharing your journey as a technologist with us. Well, thank you so much for having me.

This is a fun conversation. I really appreciate it. If you like what you've heard on this episode, please support us by subscribing to the podcast so you can get great new stories and conversations every two weeks. And please buy the book that was based on this podcast. How the Internet Happened from Netscape to the iPhone by me, Brian McCullough. Order it now wherever books are sold. How the Internet Happen. And if you weren't aware, I host a daily Tech News podcast every weekday that comes out at 5 p.m.

In that show, I tell you what happened that day in the world of tech. It's only 15 to 20 minutes long, and it's great if you love Tech News. Search your podcast app for Ride Home to find the show. It's called the TechMeme Ride Home. Thanks.