Technology
•
Jun 28, 2024
Gmail at 20
Talking to Paul Buchheit as his biggest invention enters its third decade.
Talking to Paul Buchheit as his most famous invention enters its third decade.
Max Meyer: Thanks for doing this, Paul. The first thing I want to ask is this: when did you join Google and what were your early years there like?
Paul Buchheit: I joined Google in June of 1999. I was the twenty-third employee, and we were at the office in Palo Alto on University Avenue. So it was right around the time of the Series A.
Prior to that, I was working at Intel. I had graduated in 1998 from Case Western Reserve University with a degree in computer science. It was in a rotation program, and on my second rotation, I was in the Mobile Marketing Group where I was working in the same group as Susan Wojcicki, in whose garage Google was being incubated, before the University Avenue office. I was in the same group at Intel as the landlord of Google, but didn’t realize it until we both showed up at Google.
Max Meyer: What an amazing coincidence! What was it like working at Intel?
Paul Buchheit: I didn't love Intel. It was a big, boring company. It was just very gray. I would go to work and I'd be there and feel kind of tired. I think, maybe I'll go home and take a nap, and then I go home and I'm not tired anymore. There was just no enthusiasm and I remember thinking, “man, I don't know how long I can do this, this does not give me life.”
I had always been interested in startups, but the 1990s were a very different era from today. There wasn't even a clear way of finding startups to work at. My main insight was that startups were kind of located in Silicon Valley. I took the job at Intel because it was located in Silicon Valley. So it's like Newton's method of approximation: you just go to the next step and then reevaluate.
Max Meyer: How was it that you ended up at Google in the first place?
Paul Buchheit: I was very into Linux. I started using Linux in 1993 when it just barely worked, and was a little of an open source Linux Zealot. It may be hard to appreciate that now, but at the time Microsoft called Linux “open sores.” There was a lot of fear around Linux. Microsoft would try to frighten people that if you let Linux into your organization, it's like an infection and you'll lose all your intellectual property. They really tried to scare people away from it. But I’m a very libertarian leaning, freedom oriented, market oriented sort of person. And so to me, open source was actually a really great embodiment of that. Open-source is anti-monopolistic in that you can always just fork it if someone becomes abusive.
I was interested in startups and I wanted to work in Linux, because that was the thing I was into at the time. So those were my two selection criteria. I only found six startups that were Linux based, and Google was one of them. And I used to read about Google on Slashdot.
I read Slashdot all the time. There's nothing quite like it anymore because Hacker News is unfortunately just very grumpy people. Part of what was cool about Slashdot was that they were these tech enthusiasts, genuinely excited about technology and excited about Linux and excited about open source and whatever the cool new thing was. It had that enthusiast edge. So they would cover Linux. I mean they would cover Google because Google was built on top of Linux and they were doing clusters of Linux machines. That was the thing that people loved more than anything on Slashdot, because what's better than one Linux machine? A large cluster of Linux machines!
I ended applying to Google and the five other startups, but I never heard back from most of them. Google was the only offer I got.
I actually never heard back from most of them. I got an interview with Google and with one other company, but the other company, I think we mutually didn't like each other because they never called me and I never called them. They were just lame; they didn't have good energy. But Google was cool. Google was the only offer I got.
They were smart people. You can tell in an interview because of the questions. And so, one thing I sometimes tell people who are new out of school is that when you're interviewing for a job, you're interviewing them as well. You can get a lot of insight just based on the kind of questions that they ask you as to whether the people you're working with are actually intelligent. They asked great questions.
Max Meyer: Do you remember anything they asked in the interview?
Paul Buchheit: I remember one question. Urs Holzle asked me, “Suppose you have a server and it's just running slow for some reason. How do you debug and figure out why it's slow?” The right answer is that you need to dig into the systems stats and figure out what the constraining resource is. A lot of college graduates have no idea how computer systems really work, but I loved hacking Linux so that was right up my alley!
I joined in June 1999 and it just had an energy, unlike Intel, where I could go to work and I feel energized and to me, that's the real test of a business. If a place feels alive and electric, I think that's such a strong indicator because people are excited to be there and enthusiastic about what they're doing.
It was a lively place where the level of ambition was crazy. Larry always had a huge vision for the company and had all the things we were going to do. Most people in the world encourage you to decrease your ambition: why don’t you pick a smaller idea? It was the opposite of that at Google: why are you thinking so small? Larry would always be ahead of where things were.
This is part of what made Google so different at the time. Established expert thinking on how you run an internet companies dictated that you should buy these high-end servers from Sun or HP, and you should get really high reliability hardware and software and pay a lot of money in overhead. Google sort of flipped that and said “how about if we just buy the cheapest computers we can and then we build reliability and software?” That was part of what made it such a technically interesting place; instead of trying to build these systems using expensive, high-reliability hardware, we were like let's just use a large volume of low cost hardware and then make up for it in software.
Max Meyer: That's great. I love that you were into open source stuff. What were the projects that you worked on in your first five years at Google before the Genesis of Gmail?
Paul Buchheit: I always had a lot of side projects. One of the first things I shipped was the “did you mean” feature? While I was working on the product search, we'd read a lot of query logs. I've never been very good at spelling. My brain doesn't remember information that it thinks is arbitrary. I shouldn't have to remember how words are spelled. Then I also noticed, because we work on query logs, it turns out I'm not the only one who can't spell. At least a quarter of our queries were misspelled, and I thought, why don't we just put in a spell corrector? So I launched the original “did you mean” feature to correct your spelling? The one I built I used an existing spell corrector and I tried to just filter out the really bad suggestions. A lot of words are not in the English dictionary at all, like TurboTax, for example. SoI was working on improving the spell corrector, and while I was trying to think about how to do that I was using it as an interview question. It turned out to be a really great interview question because most engineers have no idea how to build a good spell corrector and they're not even very good at thinking about it.
But there was this one guy I interviewed, Noam Shazeer, and he was already ahead of me, even though I’d been thinking about it for a while. So we hired him towards the end of the year in 2000. I was about to go visit my family for the holidays and he had just started, so for his starter project, I gave him all my code, and when I came back two weeks later, he had created the spell corrector you know now. It was revolutionary, so much better than anything that existed before, and he created that in his first two weeks on the job. What makes it a funny story is that he's one of the key people who basically invented AI, with his famous paper, “Attention Is All You Need.” Coincidentally, we were both in math league at the same time back in the nineties. I only made it to the regional math league, and he made it to the international one. But anyway, his first two weeks on the job is when he makes the world's best spell corrector, and then 17 years later invents AI as well.
Max Meyer: In the beginning, Gmail sort of started out as a search tool for people's emails at Google, is that right?
Paul Buchheit: I mean so it started out in the sense that that's what I built first, but it was always intended to be an email client. My interest in building this kind of predates Google. I was an intern at Microsoft in 1995, and at the end of that summer when I was getting ready to go home, I had printed out some white papers where I found Java. The Java programming language was created by Sun Microsystems in the early nineties for this project they called Oak, which was going to be this weird little device which was a web browser that would download applets that were written in Java.
So the really cool thing about Java was that it was written so that it could be run inside of a sandbox, so you could actually have a web browser that would download code and execute it locally, which was mind blowing. So I got very obsessed with this idea of mobile code, code that could travel over the internet, because that was something new. I just started thinking of all these great things that you could do. For example, back in this era, software only ran on your computer. When students wanted to check their email or something, they would have to go back to their dorm room, because the email lived on their computer, which was a big chunky thing, and you'd run a Windows program that would access the email. I had this idea that in the future, everything would just live on the internet and that your local device would just be a stateless cache.
So the first version of Gmail was based on a prior project I had been working on called Google Groups. I just took this indexer that I had been using and shoved all of my emails into it and that was like version zero of Gmail. I emailed it out to the engineering group, telling them I built a simple email search function.
The replies I got suggested it's kind of useful, but it would be better with their emails too, which was a feature request. So version two, I made it so that it would go scrape everyone else's emails and I launched that. Then they asked for a reply function, so I added that. So, it was an extremely iterative process where I was constantly launching something new, and continuing to improve the product.
Max Meyer: I assume there weren’t internal obstacles to building Gmail?
Paul Buchheit: No, there were endless obstacles. A lot of people inside the company thought it was a distraction because Google was a web search engine and they would ask why we were doing email? There was this notion that we shouldn't get distracted because it wasn't anything to do with web search. I started using a lot of JavaScript too, which at the time was very frowned upon at Google. The language was considered very low class and people thought you can't write real software in JavaScript. Basically, there was resistance to everything that made Gmail unique. I think anytime you do something innovative you have to expect resistance.
Max Meyer: So, we’re at the 20 year mark, and Gmail is in many ways the same product. It’s certainly bigger, and better in many functional ways, but it looks a lot like it did back then. The first thing I want to know is, do you personally still use Gmail for your emails? What are the sorts of features that you are excited about, or that you think could be improved, especially with this new paradigm of AI?
Paul Buchheit: Yeah, I use Gmail. I'm not working a hundred percent of the time, so I don't have as bad of an email problem as a lot of people who might be full-time venture capitalists. Some people are really into Superhuman and things like that, but Gmail works great for me.
Part of the vision for what comes next with Gmail was a smart assistant that does the work for you, and that’s embedded into a lot of the design decisions. Of course, it was kind of limited by the fact that Google had not yet invented AI, but the plan for Google was always to invent AI, which is why it's hilarious that they missed the boat. 20 years ago everyone at Google understood it was an AI company, but then they forgot. I think now that AI is actually here, the next step is to bring in those smart features. When some startup sends me a whole pile of legal documents, wouldn't it be great if it actually just read the documents and gave me a nice little summary, and I could have a conversation with Gmail about the emails? If you start thinking about it as a potentially superintelligent assistant who can start doing more of this work for you and even start doing tasks for you, like filling out forms. I don't know when they're gonna launch it, but it's obviously the next thing that should be done.
Assuming they do it right, adding in the smart assistant will be the biggest upgrade ever to Gmail. Of course, the part that's scary is that they can then embed thought control where it's like, “I'm sorry, I can't help you express that opinion.” The potential for just absolute narrative enforcement and thought control is frightening.
Max Meyer: What is your biggest reflection on the last 20 years and the fact that this product that you created is now so prolific? It's got to be mind blowing.
Paul Buchheit: Yeah, it's pretty cool. It's kind of a crazy experience right to be able to see something go from an unlikely, implausible vision where people say it isn’t going to work, and then it happens, and just becomes sort of conventional wisdom. It’s the same thing with Google in the early days, it's just this weird little thing and people are like that doesn't make any sense. It just creates that sense of possibility because it's not just abstract that something can go from this crazy fringe idea to something that has billions of users and is just part of everyone's life. It's something I try to pass along to startups and young students, and anyone else who has slightly crazy ambitious ideas
Technology
•
Jun 28, 2024
Gmail at 20
Talking to Paul Buchheit as his biggest invention enters its third decade.
Talking to Paul Buchheit as his most famous invention enters its third decade.
Max Meyer: Thanks for doing this, Paul. The first thing I want to ask is this: when did you join Google and what were your early years there like?
Paul Buchheit: I joined Google in June of 1999. I was the twenty-third employee, and we were at the office in Palo Alto on University Avenue. So it was right around the time of the Series A.
Prior to that, I was working at Intel. I had graduated in 1998 from Case Western Reserve University with a degree in computer science. It was in a rotation program, and on my second rotation, I was in the Mobile Marketing Group where I was working in the same group as Susan Wojcicki, in whose garage Google was being incubated, before the University Avenue office. I was in the same group at Intel as the landlord of Google, but didn’t realize it until we both showed up at Google.
Max Meyer: What an amazing coincidence! What was it like working at Intel?
Paul Buchheit: I didn't love Intel. It was a big, boring company. It was just very gray. I would go to work and I'd be there and feel kind of tired. I think, maybe I'll go home and take a nap, and then I go home and I'm not tired anymore. There was just no enthusiasm and I remember thinking, “man, I don't know how long I can do this, this does not give me life.”
I had always been interested in startups, but the 1990s were a very different era from today. There wasn't even a clear way of finding startups to work at. My main insight was that startups were kind of located in Silicon Valley. I took the job at Intel because it was located in Silicon Valley. So it's like Newton's method of approximation: you just go to the next step and then reevaluate.
Max Meyer: How was it that you ended up at Google in the first place?
Paul Buchheit: I was very into Linux. I started using Linux in 1993 when it just barely worked, and was a little of an open source Linux Zealot. It may be hard to appreciate that now, but at the time Microsoft called Linux “open sores.” There was a lot of fear around Linux. Microsoft would try to frighten people that if you let Linux into your organization, it's like an infection and you'll lose all your intellectual property. They really tried to scare people away from it. But I’m a very libertarian leaning, freedom oriented, market oriented sort of person. And so to me, open source was actually a really great embodiment of that. Open-source is anti-monopolistic in that you can always just fork it if someone becomes abusive.
I was interested in startups and I wanted to work in Linux, because that was the thing I was into at the time. So those were my two selection criteria. I only found six startups that were Linux based, and Google was one of them. And I used to read about Google on Slashdot.
I read Slashdot all the time. There's nothing quite like it anymore because Hacker News is unfortunately just very grumpy people. Part of what was cool about Slashdot was that they were these tech enthusiasts, genuinely excited about technology and excited about Linux and excited about open source and whatever the cool new thing was. It had that enthusiast edge. So they would cover Linux. I mean they would cover Google because Google was built on top of Linux and they were doing clusters of Linux machines. That was the thing that people loved more than anything on Slashdot, because what's better than one Linux machine? A large cluster of Linux machines!
I ended applying to Google and the five other startups, but I never heard back from most of them. Google was the only offer I got.
I actually never heard back from most of them. I got an interview with Google and with one other company, but the other company, I think we mutually didn't like each other because they never called me and I never called them. They were just lame; they didn't have good energy. But Google was cool. Google was the only offer I got.
They were smart people. You can tell in an interview because of the questions. And so, one thing I sometimes tell people who are new out of school is that when you're interviewing for a job, you're interviewing them as well. You can get a lot of insight just based on the kind of questions that they ask you as to whether the people you're working with are actually intelligent. They asked great questions.
Max Meyer: Do you remember anything they asked in the interview?
Paul Buchheit: I remember one question. Urs Holzle asked me, “Suppose you have a server and it's just running slow for some reason. How do you debug and figure out why it's slow?” The right answer is that you need to dig into the systems stats and figure out what the constraining resource is. A lot of college graduates have no idea how computer systems really work, but I loved hacking Linux so that was right up my alley!
I joined in June 1999 and it just had an energy, unlike Intel, where I could go to work and I feel energized and to me, that's the real test of a business. If a place feels alive and electric, I think that's such a strong indicator because people are excited to be there and enthusiastic about what they're doing.
It was a lively place where the level of ambition was crazy. Larry always had a huge vision for the company and had all the things we were going to do. Most people in the world encourage you to decrease your ambition: why don’t you pick a smaller idea? It was the opposite of that at Google: why are you thinking so small? Larry would always be ahead of where things were.
This is part of what made Google so different at the time. Established expert thinking on how you run an internet companies dictated that you should buy these high-end servers from Sun or HP, and you should get really high reliability hardware and software and pay a lot of money in overhead. Google sort of flipped that and said “how about if we just buy the cheapest computers we can and then we build reliability and software?” That was part of what made it such a technically interesting place; instead of trying to build these systems using expensive, high-reliability hardware, we were like let's just use a large volume of low cost hardware and then make up for it in software.
Max Meyer: That's great. I love that you were into open source stuff. What were the projects that you worked on in your first five years at Google before the Genesis of Gmail?
Paul Buchheit: I always had a lot of side projects. One of the first things I shipped was the “did you mean” feature? While I was working on the product search, we'd read a lot of query logs. I've never been very good at spelling. My brain doesn't remember information that it thinks is arbitrary. I shouldn't have to remember how words are spelled. Then I also noticed, because we work on query logs, it turns out I'm not the only one who can't spell. At least a quarter of our queries were misspelled, and I thought, why don't we just put in a spell corrector? So I launched the original “did you mean” feature to correct your spelling? The one I built I used an existing spell corrector and I tried to just filter out the really bad suggestions. A lot of words are not in the English dictionary at all, like TurboTax, for example. SoI was working on improving the spell corrector, and while I was trying to think about how to do that I was using it as an interview question. It turned out to be a really great interview question because most engineers have no idea how to build a good spell corrector and they're not even very good at thinking about it.
But there was this one guy I interviewed, Noam Shazeer, and he was already ahead of me, even though I’d been thinking about it for a while. So we hired him towards the end of the year in 2000. I was about to go visit my family for the holidays and he had just started, so for his starter project, I gave him all my code, and when I came back two weeks later, he had created the spell corrector you know now. It was revolutionary, so much better than anything that existed before, and he created that in his first two weeks on the job. What makes it a funny story is that he's one of the key people who basically invented AI, with his famous paper, “Attention Is All You Need.” Coincidentally, we were both in math league at the same time back in the nineties. I only made it to the regional math league, and he made it to the international one. But anyway, his first two weeks on the job is when he makes the world's best spell corrector, and then 17 years later invents AI as well.
Max Meyer: In the beginning, Gmail sort of started out as a search tool for people's emails at Google, is that right?
Paul Buchheit: I mean so it started out in the sense that that's what I built first, but it was always intended to be an email client. My interest in building this kind of predates Google. I was an intern at Microsoft in 1995, and at the end of that summer when I was getting ready to go home, I had printed out some white papers where I found Java. The Java programming language was created by Sun Microsystems in the early nineties for this project they called Oak, which was going to be this weird little device which was a web browser that would download applets that were written in Java.
So the really cool thing about Java was that it was written so that it could be run inside of a sandbox, so you could actually have a web browser that would download code and execute it locally, which was mind blowing. So I got very obsessed with this idea of mobile code, code that could travel over the internet, because that was something new. I just started thinking of all these great things that you could do. For example, back in this era, software only ran on your computer. When students wanted to check their email or something, they would have to go back to their dorm room, because the email lived on their computer, which was a big chunky thing, and you'd run a Windows program that would access the email. I had this idea that in the future, everything would just live on the internet and that your local device would just be a stateless cache.
So the first version of Gmail was based on a prior project I had been working on called Google Groups. I just took this indexer that I had been using and shoved all of my emails into it and that was like version zero of Gmail. I emailed it out to the engineering group, telling them I built a simple email search function.
The replies I got suggested it's kind of useful, but it would be better with their emails too, which was a feature request. So version two, I made it so that it would go scrape everyone else's emails and I launched that. Then they asked for a reply function, so I added that. So, it was an extremely iterative process where I was constantly launching something new, and continuing to improve the product.
Max Meyer: I assume there weren’t internal obstacles to building Gmail?
Paul Buchheit: No, there were endless obstacles. A lot of people inside the company thought it was a distraction because Google was a web search engine and they would ask why we were doing email? There was this notion that we shouldn't get distracted because it wasn't anything to do with web search. I started using a lot of JavaScript too, which at the time was very frowned upon at Google. The language was considered very low class and people thought you can't write real software in JavaScript. Basically, there was resistance to everything that made Gmail unique. I think anytime you do something innovative you have to expect resistance.
Max Meyer: So, we’re at the 20 year mark, and Gmail is in many ways the same product. It’s certainly bigger, and better in many functional ways, but it looks a lot like it did back then. The first thing I want to know is, do you personally still use Gmail for your emails? What are the sorts of features that you are excited about, or that you think could be improved, especially with this new paradigm of AI?
Paul Buchheit: Yeah, I use Gmail. I'm not working a hundred percent of the time, so I don't have as bad of an email problem as a lot of people who might be full-time venture capitalists. Some people are really into Superhuman and things like that, but Gmail works great for me.
Part of the vision for what comes next with Gmail was a smart assistant that does the work for you, and that’s embedded into a lot of the design decisions. Of course, it was kind of limited by the fact that Google had not yet invented AI, but the plan for Google was always to invent AI, which is why it's hilarious that they missed the boat. 20 years ago everyone at Google understood it was an AI company, but then they forgot. I think now that AI is actually here, the next step is to bring in those smart features. When some startup sends me a whole pile of legal documents, wouldn't it be great if it actually just read the documents and gave me a nice little summary, and I could have a conversation with Gmail about the emails? If you start thinking about it as a potentially superintelligent assistant who can start doing more of this work for you and even start doing tasks for you, like filling out forms. I don't know when they're gonna launch it, but it's obviously the next thing that should be done.
Assuming they do it right, adding in the smart assistant will be the biggest upgrade ever to Gmail. Of course, the part that's scary is that they can then embed thought control where it's like, “I'm sorry, I can't help you express that opinion.” The potential for just absolute narrative enforcement and thought control is frightening.
Max Meyer: What is your biggest reflection on the last 20 years and the fact that this product that you created is now so prolific? It's got to be mind blowing.
Paul Buchheit: Yeah, it's pretty cool. It's kind of a crazy experience right to be able to see something go from an unlikely, implausible vision where people say it isn’t going to work, and then it happens, and just becomes sort of conventional wisdom. It’s the same thing with Google in the early days, it's just this weird little thing and people are like that doesn't make any sense. It just creates that sense of possibility because it's not just abstract that something can go from this crazy fringe idea to something that has billions of users and is just part of everyone's life. It's something I try to pass along to startups and young students, and anyone else who has slightly crazy ambitious ideas
Technology
•
Jun 28, 2024
Gmail at 20
Talking to Paul Buchheit as his biggest invention enters its third decade.
Talking to Paul Buchheit as his most famous invention enters its third decade.
Max Meyer: Thanks for doing this, Paul. The first thing I want to ask is this: when did you join Google and what were your early years there like?
Paul Buchheit: I joined Google in June of 1999. I was the twenty-third employee, and we were at the office in Palo Alto on University Avenue. So it was right around the time of the Series A.
Prior to that, I was working at Intel. I had graduated in 1998 from Case Western Reserve University with a degree in computer science. It was in a rotation program, and on my second rotation, I was in the Mobile Marketing Group where I was working in the same group as Susan Wojcicki, in whose garage Google was being incubated, before the University Avenue office. I was in the same group at Intel as the landlord of Google, but didn’t realize it until we both showed up at Google.
Max Meyer: What an amazing coincidence! What was it like working at Intel?
Paul Buchheit: I didn't love Intel. It was a big, boring company. It was just very gray. I would go to work and I'd be there and feel kind of tired. I think, maybe I'll go home and take a nap, and then I go home and I'm not tired anymore. There was just no enthusiasm and I remember thinking, “man, I don't know how long I can do this, this does not give me life.”
I had always been interested in startups, but the 1990s were a very different era from today. There wasn't even a clear way of finding startups to work at. My main insight was that startups were kind of located in Silicon Valley. I took the job at Intel because it was located in Silicon Valley. So it's like Newton's method of approximation: you just go to the next step and then reevaluate.
Max Meyer: How was it that you ended up at Google in the first place?
Paul Buchheit: I was very into Linux. I started using Linux in 1993 when it just barely worked, and was a little of an open source Linux Zealot. It may be hard to appreciate that now, but at the time Microsoft called Linux “open sores.” There was a lot of fear around Linux. Microsoft would try to frighten people that if you let Linux into your organization, it's like an infection and you'll lose all your intellectual property. They really tried to scare people away from it. But I’m a very libertarian leaning, freedom oriented, market oriented sort of person. And so to me, open source was actually a really great embodiment of that. Open-source is anti-monopolistic in that you can always just fork it if someone becomes abusive.
I was interested in startups and I wanted to work in Linux, because that was the thing I was into at the time. So those were my two selection criteria. I only found six startups that were Linux based, and Google was one of them. And I used to read about Google on Slashdot.
I read Slashdot all the time. There's nothing quite like it anymore because Hacker News is unfortunately just very grumpy people. Part of what was cool about Slashdot was that they were these tech enthusiasts, genuinely excited about technology and excited about Linux and excited about open source and whatever the cool new thing was. It had that enthusiast edge. So they would cover Linux. I mean they would cover Google because Google was built on top of Linux and they were doing clusters of Linux machines. That was the thing that people loved more than anything on Slashdot, because what's better than one Linux machine? A large cluster of Linux machines!
I ended applying to Google and the five other startups, but I never heard back from most of them. Google was the only offer I got.
I actually never heard back from most of them. I got an interview with Google and with one other company, but the other company, I think we mutually didn't like each other because they never called me and I never called them. They were just lame; they didn't have good energy. But Google was cool. Google was the only offer I got.
They were smart people. You can tell in an interview because of the questions. And so, one thing I sometimes tell people who are new out of school is that when you're interviewing for a job, you're interviewing them as well. You can get a lot of insight just based on the kind of questions that they ask you as to whether the people you're working with are actually intelligent. They asked great questions.
Max Meyer: Do you remember anything they asked in the interview?
Paul Buchheit: I remember one question. Urs Holzle asked me, “Suppose you have a server and it's just running slow for some reason. How do you debug and figure out why it's slow?” The right answer is that you need to dig into the systems stats and figure out what the constraining resource is. A lot of college graduates have no idea how computer systems really work, but I loved hacking Linux so that was right up my alley!
I joined in June 1999 and it just had an energy, unlike Intel, where I could go to work and I feel energized and to me, that's the real test of a business. If a place feels alive and electric, I think that's such a strong indicator because people are excited to be there and enthusiastic about what they're doing.
It was a lively place where the level of ambition was crazy. Larry always had a huge vision for the company and had all the things we were going to do. Most people in the world encourage you to decrease your ambition: why don’t you pick a smaller idea? It was the opposite of that at Google: why are you thinking so small? Larry would always be ahead of where things were.
This is part of what made Google so different at the time. Established expert thinking on how you run an internet companies dictated that you should buy these high-end servers from Sun or HP, and you should get really high reliability hardware and software and pay a lot of money in overhead. Google sort of flipped that and said “how about if we just buy the cheapest computers we can and then we build reliability and software?” That was part of what made it such a technically interesting place; instead of trying to build these systems using expensive, high-reliability hardware, we were like let's just use a large volume of low cost hardware and then make up for it in software.
Max Meyer: That's great. I love that you were into open source stuff. What were the projects that you worked on in your first five years at Google before the Genesis of Gmail?
Paul Buchheit: I always had a lot of side projects. One of the first things I shipped was the “did you mean” feature? While I was working on the product search, we'd read a lot of query logs. I've never been very good at spelling. My brain doesn't remember information that it thinks is arbitrary. I shouldn't have to remember how words are spelled. Then I also noticed, because we work on query logs, it turns out I'm not the only one who can't spell. At least a quarter of our queries were misspelled, and I thought, why don't we just put in a spell corrector? So I launched the original “did you mean” feature to correct your spelling? The one I built I used an existing spell corrector and I tried to just filter out the really bad suggestions. A lot of words are not in the English dictionary at all, like TurboTax, for example. SoI was working on improving the spell corrector, and while I was trying to think about how to do that I was using it as an interview question. It turned out to be a really great interview question because most engineers have no idea how to build a good spell corrector and they're not even very good at thinking about it.
But there was this one guy I interviewed, Noam Shazeer, and he was already ahead of me, even though I’d been thinking about it for a while. So we hired him towards the end of the year in 2000. I was about to go visit my family for the holidays and he had just started, so for his starter project, I gave him all my code, and when I came back two weeks later, he had created the spell corrector you know now. It was revolutionary, so much better than anything that existed before, and he created that in his first two weeks on the job. What makes it a funny story is that he's one of the key people who basically invented AI, with his famous paper, “Attention Is All You Need.” Coincidentally, we were both in math league at the same time back in the nineties. I only made it to the regional math league, and he made it to the international one. But anyway, his first two weeks on the job is when he makes the world's best spell corrector, and then 17 years later invents AI as well.
Max Meyer: In the beginning, Gmail sort of started out as a search tool for people's emails at Google, is that right?
Paul Buchheit: I mean so it started out in the sense that that's what I built first, but it was always intended to be an email client. My interest in building this kind of predates Google. I was an intern at Microsoft in 1995, and at the end of that summer when I was getting ready to go home, I had printed out some white papers where I found Java. The Java programming language was created by Sun Microsystems in the early nineties for this project they called Oak, which was going to be this weird little device which was a web browser that would download applets that were written in Java.
So the really cool thing about Java was that it was written so that it could be run inside of a sandbox, so you could actually have a web browser that would download code and execute it locally, which was mind blowing. So I got very obsessed with this idea of mobile code, code that could travel over the internet, because that was something new. I just started thinking of all these great things that you could do. For example, back in this era, software only ran on your computer. When students wanted to check their email or something, they would have to go back to their dorm room, because the email lived on their computer, which was a big chunky thing, and you'd run a Windows program that would access the email. I had this idea that in the future, everything would just live on the internet and that your local device would just be a stateless cache.
So the first version of Gmail was based on a prior project I had been working on called Google Groups. I just took this indexer that I had been using and shoved all of my emails into it and that was like version zero of Gmail. I emailed it out to the engineering group, telling them I built a simple email search function.
The replies I got suggested it's kind of useful, but it would be better with their emails too, which was a feature request. So version two, I made it so that it would go scrape everyone else's emails and I launched that. Then they asked for a reply function, so I added that. So, it was an extremely iterative process where I was constantly launching something new, and continuing to improve the product.
Max Meyer: I assume there weren’t internal obstacles to building Gmail?
Paul Buchheit: No, there were endless obstacles. A lot of people inside the company thought it was a distraction because Google was a web search engine and they would ask why we were doing email? There was this notion that we shouldn't get distracted because it wasn't anything to do with web search. I started using a lot of JavaScript too, which at the time was very frowned upon at Google. The language was considered very low class and people thought you can't write real software in JavaScript. Basically, there was resistance to everything that made Gmail unique. I think anytime you do something innovative you have to expect resistance.
Max Meyer: So, we’re at the 20 year mark, and Gmail is in many ways the same product. It’s certainly bigger, and better in many functional ways, but it looks a lot like it did back then. The first thing I want to know is, do you personally still use Gmail for your emails? What are the sorts of features that you are excited about, or that you think could be improved, especially with this new paradigm of AI?
Paul Buchheit: Yeah, I use Gmail. I'm not working a hundred percent of the time, so I don't have as bad of an email problem as a lot of people who might be full-time venture capitalists. Some people are really into Superhuman and things like that, but Gmail works great for me.
Part of the vision for what comes next with Gmail was a smart assistant that does the work for you, and that’s embedded into a lot of the design decisions. Of course, it was kind of limited by the fact that Google had not yet invented AI, but the plan for Google was always to invent AI, which is why it's hilarious that they missed the boat. 20 years ago everyone at Google understood it was an AI company, but then they forgot. I think now that AI is actually here, the next step is to bring in those smart features. When some startup sends me a whole pile of legal documents, wouldn't it be great if it actually just read the documents and gave me a nice little summary, and I could have a conversation with Gmail about the emails? If you start thinking about it as a potentially superintelligent assistant who can start doing more of this work for you and even start doing tasks for you, like filling out forms. I don't know when they're gonna launch it, but it's obviously the next thing that should be done.
Assuming they do it right, adding in the smart assistant will be the biggest upgrade ever to Gmail. Of course, the part that's scary is that they can then embed thought control where it's like, “I'm sorry, I can't help you express that opinion.” The potential for just absolute narrative enforcement and thought control is frightening.
Max Meyer: What is your biggest reflection on the last 20 years and the fact that this product that you created is now so prolific? It's got to be mind blowing.
Paul Buchheit: Yeah, it's pretty cool. It's kind of a crazy experience right to be able to see something go from an unlikely, implausible vision where people say it isn’t going to work, and then it happens, and just becomes sort of conventional wisdom. It’s the same thing with Google in the early days, it's just this weird little thing and people are like that doesn't make any sense. It just creates that sense of possibility because it's not just abstract that something can go from this crazy fringe idea to something that has billions of users and is just part of everyone's life. It's something I try to pass along to startups and young students, and anyone else who has slightly crazy ambitious ideas
About the Author
Maxwell Meyer is the founder and Editor of Arena Magazine, and President of the Intergalactic Media Corporation of America. He graduated from Stanford University with a degree in geophysics.
You might also like
