Bharat Mediratta - CTO at Dropbox

(Joel Beasley at 00:00:00) Hello, my friends. Today we are talking to Bart, the CTO at Dropbox, and we discuss how to explore new technologies as a CTO, ways to think about site and product reliability, and a framework for acquisitions. All of this right here, right now on the Modern CTO podcast. Here we go. This is the Modern CTO podcast.

(Joel Beasley at 00:00:31) Hello. Hello.

(Bart at 00:00:33) Hey, Joel.

(Joel Beasley at 00:00:34) Hey. Your camera looks awesome. What type of camera is that?

(Bart at 00:00:37) You know, I decided with COVID to invest, and I started looking around, and I found that you can just take a DSLR and use it as your webcam. And when you do that, you get much better image quality because in this world, connecting with humans is hard. So the more you can show your facial expressions, the better. You know? This allows me to zoom in a little bit or control a little bit the image, which makes a huge difference, I find.

(Joel Beasley at 00:01:06) What camera did you go with?

(Bart at 00:01:08) You know, I just had this old Canon Rebel T5i that I was using back when I was really trying to learn photography. And, you know, there's software now that works with a lot of these just turns it into a virtual webcam. So it's nothing special. Yeah. I think it's even the stock lens. I mean, this is actually cheaper than a lot of the higher end webcams. It's trivial.

(Joel Beasley at 00:01:29) Yeah. It looks great. I think we have a Canon T7i. And I tried to use an older camera too, but the problem was when doing the live stream, there was these white boxes on the output for the HDMI. So I had to upgrade the camera to be able to turn that off.

(Bart at 00:01:47) Got it. Got it. Yeah. I'm not trying to professionally produce my videos. A little bit easier than you, but you've got a very nice microphone right there. Maybe I should step up my microphone game.

(Joel Beasley at 00:01:59) Thank you. Yeah. It looks so official. Right? So you've got the Google arm and everything.

(Bart at 00:02:05) That's weird.

(Joel Beasley at 00:02:06) We try. We try. Actually, so I'm very excited to talk with you because I'm a huge fan of Dropbox. We've been using Dropbox for all of our content since day one. We have, I think, ten terabytes of video and audio and different stuff in there, but I'm we're just huge fans of the product.

(Bart at 00:02:23) Fantastic. I got to ask, what do you like about it?

(Joel Beasley at 00:02:27) It works. I would say that would be the most simple thing. When we needed this solution to store stuff digitally and, you know, we have our producer and editors and everybody needs access to everything, and we had to figure out our workflow, educate ourselves, but it just works. And then I guess the other cool thing is that it's got this feature to recover files, like if you mess up or delete something or something happens.

(Bart at 00:02:58) Yeah.

(Joel Beasley at 00:02:59) So you can recover files. So I think those two things are the big things that we noticed. It's interesting though because if you think about it, some of your best tools that just work and solve the problem, you don't think about them much.

(Bart at 00:03:12) Yeah. Exactly. That's exactly right. You know, this is what we thought about at Google when we were doing search. It's like you just want it to work because the things that have really become utilities in your life are the things that you don't have to think about. They're like the air you breathe. You know? And then you wind up using them more and more, and they become structural in your life. So that's great. That's the best feedback that it just works. Especially because for my domain, my job is to make sure that it just works. You know, aside from the product experience, is it massively reliable, and does it do exactly what it says? Does it never lose any data? You know, always make sure that we give you that experience day after day. That's how you become a utility in someone's life.

(Joel Beasley at 00:03:48) I am curious. How do you respond when people, you run into people in your everyday life because Dropbox is a household name, and they've got, you know, their two cents about product features. How do you as the head of the technology at the company, how do you respond to that?

(Bart at 00:04:05) You know, I find that everyone can teach me something. You know? And really, what they're teaching me usually in that moment is what their needs are and what their experiences are. And what I find is the easiest mistake to make is to get disconnected from your customers. So that's why I always ask. When someone says they like it, you know, I always ask, well, what do you like about it? Because, you know, if you say you didn't like it, I'll say, what don't you like about it? Like, I want to understand your experience and your perspective. I mean, you sit in a really interesting space where you're doing huge amounts of video and audio processing, and you need your data to flow through that pipeline. But there's other people who sit in the space of saying, well, you know, I just need to make sure that all of my personal files are stored conveniently. And a lot of times people have confusion. They're like, I don't understand why you make it so complicated. And really what they're saying is, I want this to be more of a utility, and you're making me think about it too much. And that's great. That's great feedback. That's feedback we could take back to the team and say, look, maybe what we should be doing is steering towards making the decision for you that's right 80% of the time and letting you reverse it instead of making you have to choose every time. I think we've historically, in my career, have always gotten great feedback from just asking people, what do you like about it? What do you not like about it? And, you know, you don't have to act on it right away, but you file it away when you think about it.

(Joel Beasley at 00:05:21) That's why, you know, when we first started talking, I was like, I'm a huge fan of Dropbox, and you all get so many things right, and I've been a customer, I think, since either 2007 or 2009. Like, it was a long time ago. But I remember your original interface, and at the time, it was the most advanced thing I'd ever seen. I was like, oh, the designers there are crazy cool. Like, it was so simple, and everyone else had that 1995 Windows look for their file folders, and you guys had this super modern look. I was just so excited about that. Yeah. So I've just been a fan since the moment I needed this type of solution. And so, yeah. Were you a fan of Dropbox before you started? What brought you to Dropbox?

(Bart at 00:06:06) Yeah. You know, I first heard about Dropbox in 2007, I think, when Drew took it to YC. And, you know, I had this problem like everyone else did. You know, I was at Google, and I had my personal network and my work network. And Google internally did a great job of file storage. You know, Google was, at that point, one of the world leaders in it. But I had this real problem that you couldn't use that technology effectively for your personal life. So when Dropbox showed up, you know, I was one of the early subscribers. I was like, this is fantastic. You know? I think I heard about Dropbox because they did this great viral marketing. You know, I was wandering around San Francisco, and someone walked by me and gave me a business card. You know, back then, business cards were the norm. And it was a weird interaction. Like, it was someone I just met. And it wasn't actually a business card. It was a coupon for, I don't know, two gigabytes of data in Dropbox or maybe not even that much. It just doesn't, it's probably 500 megabytes or something. But it was great. I was like, oh, wait. I can just go in because at that point, I knew I needed this, but I was cobbling it together. You know, I mean, I'm a techie. So I had my own FTP server posted on a machine in a data center. You know, I was running an open source project, so I had my own infrastructure. And I would store stuff. Anytime I wanted to share a file with my friends, I was always giving them these weird URLs and stuff, and there was no sharing model. And it was all security through obscurity. And I was like, god, this is just, even then, you know, I was much younger. Even then, I could see that this was not sustainable. It was like, I would have to keep spinning the plates on all my various servers to build my own file sync and share network. And I was like, please, let there be a consumer solution. And so when someone came up and gave me this business card and I started using it, I was like, wow, that's a complete game changer. You know? And so, and then I had some friends who went to Dropbox in the early days. I was very happy at Google. I was doing, you know, some really great mission work that I really enjoyed. So I wasn't seriously looking for a job. But some of my friends who I really liked and respected went to Dropbox. And so I visited offices, and I met some of the folks there. And, you know, I wound up joining Dropbox back in 2014, when I was leaving Google. I'd been there for eleven years. And at the time, Dropbox was a really interesting opportunity for me. So I sat down with Drew and Arash and other members of the team. But at the end of the day, I decided at that point that I really wanted to go do something radically different. I wanted to go apply, you know, I felt like I had built up this deep domain expertise and technology and, you know, understanding how to deploy things at scale. And I really had this sense of impact. So I wanted to do something different. So I wound up not going to Dropbox in 2014, but I advised the team. You know? So Aditya, the CTO at the time, asked me to come in and advise some of the engineering leaders. And so I did. I did that for a couple of years. And I got to know the team. And, you know, I really liked and respected what they were doing. And so when I started looking around again last year, I met up with Drew again. And I was really inspired by the vision he has, you know, for making work more human, you know, designing a better way of us working together. And frankly, with the advent of coronavirus, it's become more important than ever for us to figure out how to be human. I mean, this is one of the reasons why I get a nice camera, and I think very carefully about how you connect with other people. Because in this moment, this is the norm, being on a Zoom and all of my meetings. So as I sit in this room for ten hours a day on these meetings. So it's really, really important to figure out how to make this a more human experience. And I was really excited last year about the vision Drew had, and I'm really excited to see where we're able to take it now. You know, now the world is moving towards us in terms of the need for this is growing.

(Joel Beasley at 00:10:02) Yeah. I saw you guys were getting into different types of technology other than just file storage. Can you tell me a little bit about that?

(Bart at 00:10:10) Yeah. I mean, at the end of the day, oops. Sorry. Hang on a second. Hey, buddy. I'm in a meeting. Just one more break. Okay. But I got to do this meeting. I hope you're enjoying school. Yeah. Okay. Yeah. I hope you're enjoying your meeting. Thank you.

(Joel Beasley at 00:10:23) Oh, that was so sweet. How old?

(Bart at 00:10:28) He's five. This is his first day of school, so he's super excited.

(Joel Beasley at 00:10:33) Oh, is he being homeschooled?

(Bart at 00:10:35) Well, no. I mean, so California, school is all online right now. It's all distance learning. And so this is a really traumatic time, by the way, for parents because, you know, now you've got your kids. So over the summer months, you know, you're in a situation where you're playing dual role of trying to work and trying to take care of your kids. But they don't have a structure, which is good and bad. Right? They don't have a structure, which means you can make a structure. But now that they're going back to school and it's distance learning, they're in your home. They're using your bandwidth. Right? So you may see me drop out or my audio may go because I'm sharing my bandwidth now with a bunch of other people. And, but they have a structure. And that structure is very demanding, and they can't always do it themselves. So, you know, I've got three kids. I've got an 18-year-old who's a sophomore in college, a 17-year-old who's a senior in high school, and a 5-year-old who's just started his first day of kindergarten. And they all have completely different experiences. And you have to figure out how to make those experiences human in this world. So Remy today is getting an intro to his teachers, and they're talking about the school that he hasn't set foot in and may not for months, you know, maybe not even the whole school year. And they're trying to create this experience that you have on this first day of school where you get to show up in your classroom and meet your friends and shake hands and introduce yourself. And you go on a tour, and you're like, here's the library. Here's the cafeteria. You know? Say hi to the janitor. Say hi to the librarian. And they're doing this all on Zoom. The whole world right now is in this massive experiment. And, you know, the challenge is that our tooling was not set up for it. You know, watching the schools reorient towards having this type of tooling is incredibly hard. And so back to what you were asking about Dropbox, we believe that we can actually help with this because we sit at a really interesting place. We have, you know, we have massive amounts of users' files, and we have user trust. And we have this one-on-one relationship with you because there's no third party involved. Right? And so we built up trust and credibility in this space, and we have the infrastructure to now deliver more experiences on top of it. So what does that mean? That means that in the challenges you face today, you have to figure out, how am I going to organize my life? How am I going to get through my day? How am I going to communicate with people? What are my workflows when I can't meet in person? When some things that were paper are no longer paper. You know, look at companies that are doing real estate transactions. You know, they're very used to everything being signed on paper. You've got to move to eSignature solutions. Right? You know, we're seeing a massive surge in HelloSign, which is a company we acquired last year, doing eSignature solutions and workflow solutions, you know, digital asset management. All these areas where you need to shift the way that you work to continue working into this new environment. Same thing that's happening with school is happening in the workplace. So we are being very thoughtful about what can we do to start from that base of files to start adding more and more value to everything you do beyond things like Smart Sync, into things like collaboration, you know, into figuring out ways that you can be more productive with the tools you have in the situation. It's fun. It's a fun challenge.

(Joel Beasley at 00:14:01) Did you guys build a password manager?

(Bart at 00:14:03) We did. We built a password manager. We built Vault. These are pretty cool things. We built, we launched Dropbox Transfer. The password manager is really interesting because you want to, what are the things that a trust relationship really buys for you? Right? Like, you want to store your password somewhere that you trust. Right? You want to make sure that your files in your Dropbox are secure. It's more than just, I mean, your tax forms, your medical records, and your random correspondence are three completely different categories of things. Right? Like, you have to think about them differently. You have to reason about them. You have different security models.

(Bart at 00:14:46) And there's so much more we can do in terms of organizing your data as opposed to just giving you a folder. We're trying to move beyond this—you have a folder and you can put anything into it—to a smarter and smarter organization system for your life, and then go from there to helping you collaborate with other people in ways that make sense. If you share a folder accidentally with someone, and that folder has a subfolder in it which has all your medical records, you're in a pretty bad place. You want extra levels of thought and care and security to go into how that stuff's organized. And we're well positioned to actually start moving up that stack and sharing more and more of these capabilities with our users.

(Joel Beasley at 00:15:27) How do you decide what to build? When you guys were thinking of the password manager, was it feedback from customers, or were you laying out the landscape and saying, you know, what's next for the evolution of Dropbox and picking? How were you deciding of all the different types of tools or services to build? How did you decide password managers?

(Bart at 00:15:48) Right. I mean, so we started with our vision to design a more enlightened way of working. So that's aspirational and it's up here. We think, okay, what would be the things that would ladder up to that?

(Bart at 00:15:59) Right? It's like, okay, we've got this base of files and we want to get up here. And so we start thinking, okay, first, we need to bring more of your life onto our system. Right? It's not just PDFs. It's not just random storage. It's like, no, it's actually your structured life.

(Bart at 00:16:14) Well, what's in your structured life? Well, there's the documents that you get digitally. There's the way that you organize them. So that's things like medical forms versus financial forms versus correspondence, et cetera. You know, photos versus documents, et cetera.

(Bart at 00:16:30) And then there's also how you organize all the things in your life that live in the SaaS space. And what is the key to organizing your life in the SaaS space? It's actually a couple of different factors. One is how you get access to it. And the second is, can you allow things to index it and start understanding it?

(Bart at 00:16:46) Would you be able to say, hey, I log on to this site and I use it for a while, but then I want to leave it. But I want to take my data with me. Well, wouldn't it be great if Dropbox understood, first of all, how to create that access for you? And Dropbox can then sit at this central place of saying, I understand the web of SaaS applications that you live in.

(Bart at 00:17:08) And then you could move stuff in and out of Dropbox as you start saying, well, I'm going to use this for a while. I want to deprovision. I want to stop using it. I want them to stop having my data. I want my data to go to a trustworthy place where I can keep it and organize it for the long term.

(Bart at 00:17:24) Having passwords, having transfer, having that connectivity, being able to do cloud-based search and indexing, being able to have a central vault where you can put that data becomes a central place that you can organize your life from going forward. So these are ways that you can actually start bringing together your life. And so when you think of it that way, then access becomes important and passwords becomes a key part of access. And so Vault—you know, the Vault acquisition, the password manager we launched—becomes pretty integral. So that's just kind of the way that we think about decomposing that big vision into a series of steps we could take so we can start helping users move up the ladder.

(Joel Beasley at 00:18:07) It looked like—so you acquired a password manager?

(Bart at 00:18:10) Mm-hmm. Yeah. It's confusingly named Valt, V-A-L-T. And then we launched a product called Vault V-A-U-L-T. So basically—sorry, you'll hear me say vault, Valt, vault, Valt. It's actually—yes, we acquired V-A-L-T Valt, and then brought the team in, and they did an amazing sprint, used their kind of domain expertise and knowledge, worked with our security team, worked with our infrastructure team, rebuilt that product on top of our infrastructure, and we just launched it.

(Joel Beasley at 00:18:36) Oh, that's pretty cool. So you mentioned that you acquired HelloSign for the digital signatures. So that's smart because I was going to ask you the questions about, like, you know, is the password manager just an experiment that may go away? But it seems like you're acquiring lines of business, and they already have customer bases and revenue. And so you're just building—you're just pulling in, and these aren't things that are going to go away. They're not experiments. They're things that you were doing.

(Bart at 00:19:04) Yeah. I mean, when you think about acquisitions, you kind of—I break them into three categories. There's business, there's product, and there's capabilities. So sometimes you acquire stuff just from a business perspective. It's like, hey, you know, we see a gap in our portfolio and having that thing plugged into it makes a ton of sense. So that's one approach. The product approach is, hey, we see the jobs to be done that our customers have, and we see the broader product arc that we want. And so we're going to start acquiring products that fit in. Right? And so, typically, then you take the product, you put it on top of your infrastructure to drive efficiency, and then you wind up adding it to your product portfolio. And then there is kind of capabilities. It's like, wouldn't it be great if we had the ability to do cloud search or digital signatures or password management? And what you really want in these systems is for them to be connected together.

(Bart at 00:19:57) So you acquire a bunch of capabilities, and you put them under a platform. You know, we call it internally product platform. And then we expose them through APIs internally. So we can say, hey, any of the products we have can use any of these capabilities in a way that's well understood, operationally reasonable characteristics, reasonable cost efficiencies, essentially more than the sum of its parts.

(Bart at 00:20:21) And then your products are the same way laddering up to your business. You're saying, okay, how could we bundle or unbundle this product offering to meet different customer segment needs? And then from a business perspective, you're saying, how does this all add up to a long-term business trajectory that really makes sense? So that's how they kind of ladder up.

(Bart at 00:20:40) And then when you kind of look down on this, you're like, okay, great. How would we think about what the business needs and how we satisfy our customers to continue this valuable one-on-one relationship? So it's like, for the business needs, it's what could we produce? What does the customer need? And how are they connected? And then from there down to the product, and then down from there to the capabilities. So I tend to live at the capabilities layer. Like, what capabilities do we need to bring under our product platform? Timothy Young, our SVP and GM of the core Dropbox business, he thinks about it more in the product lens.

(Bart at 00:21:12) And then, you know, Drew and Olivia, our CEO and COO, are at the business level of, like, what do we need at the—I mean, that's rough characterization because we all do everything together. But when you think about primary responsibilities, you kind of need those three pieces working together to make sure that what you bring on winds up being additive across the business, the customers, and the technology.

(Joel Beasley at 00:21:34) Yeah. I was talking with this guy, Matt. He's the founder of this company called Gremlin about this concept that was fairly new to me called chaos engineering. Have you come across this yet?

(Bart at 00:21:46) Chaos engineering. I'm not sure I know the context of that. I mean, I know—

(Joel Beasley at 00:21:54) I'll go deeper. Right? Yeah. So from what I get, and it was a new term for me too—and so that's why I Googled and there was an O'Reilly book on it—but it's basically this concept of breaking things intentionally.

(Joel Beasley at 00:22:06) It's like you break them, you learn from them, you track your learnings.

(Bart at 00:22:10) Right. Right.

(Joel Beasley at 00:22:12) And it was really kind of—he had started, I think, at Amazon. And when he was at Amazon, his job was to get the website's fatal errors down.

(Bart at 00:22:23) Mm-hmm.

(Joel Beasley at 00:22:23) Right? And they cut them in half, and they had all these successes or whatever. But he took all the, I guess, the learnings from that and ended up building a suite of tools to help engineers do this chaos engineering. So I was just curious, like, you know, I'll give a specific example of a chaos engineering scenario. It would be that your server could triple the traffic to it, and it would be fine, or it would spin up another instance. Right? So you would make that statement. You would assert that that would happen, and then this test would run, and it would create the scenario where it would actually stress test or load test. And they have just hundreds of these different things you can test. But I was just curious how larger organizations are—how they're referring to that. Like, what are they calling it? Site reliability engineer? And then how are they testing it? So I didn't—and you might not even be the right person to ask. You may point me to someone else in the organization, but I was just curious for my own sake.

(Bart at 00:23:20) Yeah. I mean, it's a fantastic concept. I am familiar with this. I haven't heard it referred to as chaos engineering, but that totally makes sense. You know, the first I heard of it was Netflix doing it. They had a thing called Chaos Monkey, which is a process that runs inside Netflix that just creates entropy. It just randomly turns things up, shuts things down, breaks things, you know, does weird—and you're testing your resilience. Right? We do the same at Dropbox. We have a similar mechanism. It's similarly branded. And what it does is within a certain set of parameters, it's just like a bull in a China shop. It just goes and it does a bunch of random things, and we can unleash it. We have a few protected areas where we know things are a little weaker, and we're shoring them up. But it's great. It's great to find all the ways that things can go wrong. And you can learn a huge amount from that because it exposes your biases. Whenever—in all walks of life—I mean, you know, I spent five years at an EdTech startup, and I learned a lot about cognitive theory and how humans really are influenced by their biases among a great number of other learnings in startup land and education land. But one of the things that you really learn is that we all have a mindset and we have biases. And they're the air we breathe and the water we swim in, and you don't even notice it.

(Bart at 00:24:41) And so the advantage of chaos engineering is it really challenges those preconceived notions. Like, oh, that could never fail. That could never happen. You know, I got a buddy who works at SpaceX. And when they were diagnosing why one of their Falcon boosters blew up, it was because of a situation that they believed could never happen and they hadn't guarded against it.

(Bart at 00:25:01) And that's okay. Those things are going to happen. And the stakes are very high in that situation. The key is to avoid the stakes getting high in the first place. Right? Ultimately—and no knock on SpaceX, they're doing amazing work—there's always going to be some failures. But the trick is to inject enough chaos and entropy into your system as you build it so that you know. So I would say that the right word for it is building a robust architecture. Right?

(Bart at 00:25:30) Building a robust architecture requires you to challenge your mindset to understand where you might have some blind spots and find ways to uncover those. No matter how painful it is, you have to imagine what would happen. So we have our own data centers. And Dmitri, who runs our data centers, you know, he will go into a random data center every once in a while and just pull a switch. Okay. What happens if I do this? You know, when you'll ask the team, you're like, what do you think is going to happen? And then they'll form some theories. And, you know, they're generally—they're very talented. They generally know.

(Bart at 00:26:05) I'm thinking about forming theories. And then he'll test it. Okay. Great. But let me pull that. And we have automated systems that do similar things. Right? Like, what happens if this failure happens? But it's not just technology. So if you think about it in terms of team, technology, and processes.

(Bart at 00:26:23) Right? It's a team mindset. It is first acknowledging that you have biases. Even if you don't know what they are, but you know that you have them. Right? And recognizing that you have to find them and challenge them. That's step number one. And then building technology to help you find your biases and challenge them. Right? And then it's processes, making sure it's part of your ongoing culture. Right? This is something we care about, something we prioritize. At Dropbox, we do something called Wheel of Misfortune where people get together, and they run these exercises where they say, okay, you spin the wheel. It's like, oh, we've had this failure. Okay. Let's talk through what would happen in that failure, and then let's start figuring out what would go wrong and how would we fix it. And trying to—it's almost like a pre-mortem. It's like, okay, we've just suddenly lost all power to this data center. What happens? How long would it take? What's our theory about what happens? You know, prepare for it, practice it, and then try it. Right?

(Bart at 00:27:19) At some point, you know, let's not be reckless. Let's not lose our users' data or break trust. Let's not take the site down. But let's understand how we can prepare for these things. Because if you believe it can't happen and you never prepare for it and then it does happen, you don't have a robust system.

(Bart at 00:27:38) And that's, I think, you know, the real value of chaos engineering. It's part of a broader—well, I haven't read the book, so maybe the chaos engineering principles talk about it more broadly. But it is that broad set of things across the team and the tools and the processes that you need to make absolutely sure that you're building a robust system.

(Joel Beasley at 00:27:57) I like the branding around all of them because you've got the Chaos Monkey and the gremlins, that little gremlin that rips everything apart. I think it's kind of fun. It's—as an—so I've been an engineer programming for, you know, over 17 years. So I like it when there's things that you come across that are fun but also useful. It's kind of rare, but when you do come across it, I'm like, oh, this is new. You know, when you've been doing something for that long, when you come across new stuff, it's very exciting.

(Bart at 00:28:26) Yeah. Yeah. And it's a nice evolution of the process. Right? And I think we've seen it as systems go to scale. You know, Google used to say, if you take a problem and you make it an order of magnitude larger, it ceases to be the same problem. It becomes a whole new problem because of the scale you've taken on. You make it 10x larger again, it's a different problem at scale. So as we started getting to this scale, truly, this is when technologies like this are emergent and necessary.

(Joel Beasley at 00:28:55) Okay. So speaking of scale. Right? I was curious about this. So let's talk about cloud storage in general because you're very bright. And whenever I meet people like you, I'm just curious, and I want you to answer questions and help me understand things or wrap my mind around concepts. Right? So I'll go to a cloud storage site and it'll say, like, five nines. Like, we're 99.999% whatever safe or secure or whatever. It's to make me feel good.

(Joel Beasley at 00:29:21) Right? So being the nerd I was, I was like, alright, well, if you have x amount of exabytes or whatever, then you're losing hundreds of thousands of gigs, and, you know, whatever it is. I assume that that's wrong. There's some advanced system.

(Joel Beasley at 00:29:37) There's a way that... can any cloud storage provider say that we've never lost anyone's data?

(Bart at 00:29:46) Well, when they talk about all those nines, they're really kind of aggregating several different things. I'll just talk about the storage space for a second, because it's different from storage and compute, you know, network, et cetera. So just storage, just to narrow it down. Storage really has three factors. There's durability, there's reliability, and there's integrity.

(Bart at 00:30:05) Right? And they're each going to have a different set of nines associated with them. So reliability, like site reliability is, you know, how often is the site down? And reliability is important, because if you want to make it 100% reliable, don't change it ever again. That's great from a technology perspective.

(Bart at 00:30:25) SREs are like, that's fantastic. But that's bad for the business. Right? It's bad for the user. We need to advance things like Smart Sync so you have a better experience, which means we've got to push new code, which means the site might go down.

(Bart at 00:30:35) You know, that's the challenge. And you're always dealing with the challenge of the ecosystem evolving. There's newer hardware. There's new data centers. Things are evolving.

(Bart at 00:30:48) So reliability is typically, you know, three or four nines, where you think, okay, that's a real trade-off. Am I willing to take, say, five, ten minutes of downtime a month to get the velocity of being able to push faster? And I think, you know, users really do appreciate that tension. Right?

(Bart at 00:31:08) They want more features. They want things to improve. And if I went to you and I said, listen, do you mind if Dropbox is down randomly five minutes throughout this month so that we can push—we could do two pushes and ship more features to you? You'd probably take that offer, because the reality of it actually impacting you is pretty low, but the benefit is probably meaningfully appreciable. So reliability is separate from durability.

(Bart at 00:31:34) If you stick something, a file in Dropbox, how durable is it? Will it last forever? Will it ever go away? Will you be able to come back to it ten years from now and still find that same file? And I can tell you, you know the answer to that.

(Bart at 00:31:48) Right? You probably put files in there in 2007.

(Joel Beasley at 00:31:50) I did. I saw them. They're like some CSS files. I saw them right before this call. I was like, no way.

(Bart at 00:31:56) Right. So I mean, obviously, we don't have decades of data, but we have fourteen years of data, thirteen years of data. Those files are durable. And then the question is integrity. If you open up that file, is it going to be exactly the same as when you put it there?

(Bart at 00:32:13) Right? And durability is high through replication. Right? You basically say, okay, for every user gigabyte, we replicate it across multiple clusters in data centers around the world.

(Bart at 00:32:26) We shard it out into blocks. We make absolutely sure that we will never lose this file. And that has many more nines than just three or four of site reliability. And then there's integrity. Right?

(Bart at 00:32:39) Will we ever corrupt a byte in any of your data ever? And that has a very long list of nines. Right? That is something we care about the absolute most, because if it's one thing if your site isn't available for five minutes, okay.

(Bart at 00:32:54) It's another thing entirely if we lost your data or worse, we mangled your data. Right? And so for integrity, you really care about every single thing that happens. So, you know, when AMD released new chips, we detected a single bit flip in one of the calculations across billions of operations. And we went after that single bit flip to find a flaw.

(Bart at 00:33:18) I mean, in one of the AMD processors. You know, these are normal things that happen. We partnered up with them. We had some deep technical experts go in there, figure out backtracking that to the actual flaw. Because at scale, a single bit flip that happens across a billion operations is going to damage your integrity. You can't flip a bit on a byte in a file.

(Bart at 00:33:41) And it doesn't matter if you went back into that CSS file that's thirteen years old and you opened it and you saw one character out of place. That would be our failure. Right? And that's an unacceptable failure. So when you think about it, you know, just from the storage and files, I could share perspective, these are the axes that they talk about.

(Bart at 00:34:00) Now, I don't have the exact stats on hand, but our integrity number is like seven or eight nines. Right? So what does that mean? Does that mean that we will lose data from time to time, or we will have integrity issues? It doesn't mean that.

(Bart at 00:34:17) It means that the statistical likelihood of it is vanishingly small. Now we need to be eternally vigilant to make sure that everything we do prevents that from happening. And we need to make sure that we have a plan if there is a bit flip, to detect it and recover from it. Right? So, you know, no one will ever say 100%, because you can't know for sure.

(Bart at 00:34:38) So what you do is you wind up saying, how many nines am I willing to pay for, and how do I keep pushing that envelope further and further? And that's what you see kind of aggregated in a lot of the sites that are like, hey, we're 99.999. They're aggregating all these things, and they don't want to try to break it down and explain it in this kind of long, lengthy way that I just did, because the typical consumer is like, I just want to trust them and turn off my brain and not worry about it.

(Joel Beasley at 00:35:04) I feel better now, because hearing you talk, as you're talking, my brain's just like, this guy is so sharp.

(Bart at 00:35:12) Thank you. Well, listen, I've got an amazing team that does the work. So don't mistake me for the person who does the amazing work.

(Joel Beasley at 00:35:19) I don't.

(Bart at 00:35:20) I don't. Team at Dropbox.

(Joel Beasley at 00:35:21) The flattery is on your communication skills, which is what I focus on. Right? The way you can articulate your thoughts and they're clear, and you can't say these things without having thought about them at great length. It's just not possible. Right?

(Joel Beasley at 00:35:34) And for me, I respect that a great deal. I'm curious. You mentioned San Francisco earlier. Is Dropbox headquartered there?

(Bart at 00:35:45) Yeah. Our headquarters are in San Francisco.

(Joel Beasley at 00:35:47) Okay. So I have spent a lot of time out there, a good amount of time, maybe three or four times a year for ten years. Right? And I saw—I think their name's Palantir. I don't know how to pronounce it exactly.

(Joel Beasley at 00:36:00) But, you know, billion dollar technology company, and they're like, we're leaving Silicon Valley. It was in the headlines last week. Did you see that article?

(Bart at 00:36:07) I didn't, but, you know, there's certainly been a few.

(Joel Beasley at 00:36:10) Yeah. So they blamed, or they mentioned... the title was Palantir Moves to Boulder Because of Silicon Valley Monoculture. And I was like, so I Googled that phrase, and it wasn't the first time it was used. There's lots of articles about it. Can you give me some insight or thoughts on that?

(Bart at 00:36:30) Yeah. You know, I've lived in Silicon Valley since 1992. And so I've watched a few different cycles here. It's crazy, actually. Hard to believe it's been that long.

(Bart at 00:36:42) I don't know why I'd put it that way. Wow. Twenty-eight years. But, yeah, I saw pre-web. I saw Web 1.0, 2.0, 3.0.

(Bart at 00:36:51) I mean, I don't even know what web we're on now. But the interesting thing about Silicon Valley was in the seventies and eighties, it was defense. It was all Northrop Grumman and, you know, Lockheed Martin. And, you know, we had the Moffett Air Base for, you know, where NASA was doing tests. And San Francisco, it was a beautiful, small, much more blue collar culture.

(Bart at 00:37:18) San Francisco is 25% roads and sidewalks, but also 25% two-story buildings and beautiful houses, Victorians. And San Francisco has evolved over that time. So what happens when something takes root and it has legs is that it starts to squeeze out everything else. So, you know, the culture in the nineties when I came out, I joined Sun Microsystems. Sun Microsystems was a big deal back then.

(Bart at 00:37:49) But it was pre-web tech, and then web kind of moved in. And then, you know, I would say the Yahoos and the Excite@Homes were the early stage. Google really planted the flagship in the South Bay. And when Google demonstrated that you could make a massive, you know, trillion dollar empire off of this, and Apple had its resurgence, even the big companies like the biotech pharmaceuticals like Genentech, the big companies who are here started to feel the squeeze.

(Bart at 00:38:12) Because what happens? Right now, you've planted a flag and everyone's like, I want to work for Google. And this is pre-COVID, so they have to actually physically come to the Bay Area. They get a job that pays a lot of money. The company's gone public, and they're transacting that into the real estate market. They're buying real estate, and they squeeze out people who don't have this. So a lot of what happened, you know, and I'm not an expert on this, but what I've seen is that it just became less and less economically viable to have diversity, because you had this one kind of tech influence that just had so much money.

(Bart at 00:38:54) And there's a huge backlash in San Francisco against it. People miss the old days, but it's also the modern reality. I remember in 2006, Larry Page got up at a Google all-hands when people were saying, please, can we open a San Francisco office? And he said, over my dead body. We're never going to have a San Francisco office.

(Bart at 00:39:10) Right? And it was the right call because San Francisco was too close. It would siphon off the energy from Mountain View, but it was also inevitable. Right? And so now all the companies have offices in San Francisco. You know, big, huge, profitable startups in the tech space—the Ubers of the world, you know, Lyft, you know, Dropbox.

(Bart at 00:39:29) They all started in San Francisco. And so what happened is, you know, more and more you have this culture of people joining to work in tech, tech gaining more momentum, the tech money and the tech people squeezing out people who aren't in tech. And your culture, it doesn't become necessarily a monoculture, but it becomes increasingly homogenized over time. And then, of course, there's a backlash. You know, tech is not just an end in and of itself.

(Bart at 00:39:58) Tech is in many ways a means to an end. So now you have expansion into, you know, more biotech and pharmatech and, you know, medical, health tech, fintech. But still, it does create this culture, and not everyone wants to be part of it. I mean, if you're a company that wants to hire people that don't do tech, this is not the best hunting grounds anymore. And this will change.

(Bart at 00:40:26) Nothing lasts forever, but this is where we are now. And so I'm not too surprised that this is happening. The really interesting thing is what happens with coronavirus? What happens when some of these companies like Twitter says, hey, you can work from home forever? Google says you can work from home until, you know, the end of the school year.

(Bart at 00:40:43) All of a sudden, where you live is dictated not by proximity to job, it's dictated by the quality of life you want to have and the people you want to live near. Because the same way that you and I are having this call now is how I do all my meetings. I could be anywhere. I don't have to be anywhere near. In fact, I haven't been to the office in six months.

(Bart at 00:41:07) And in fact, no one has since we kind of shuttered it for coronavirus. So what's interesting to me is not where we are today, but the broader arc for where we will be in a year. Imagine that just 50% of the big tech companies in San Francisco go to this virtual mode. You know, I'd say that's probably not too far off one way or the other. It's probably not 60, probably not 40.

(Bart at 00:41:30) So call it 50%. That will change the culture, because now you can work for these tech companies without moving to San Francisco. So that will relieve housing pressure. That will relieve, you know, that will create space for the city to change and the valley to change. But, you know, it's still too soon to tell.

(Bart at 00:41:47) This is just my prognostication, you know, based on what I'm seeing.

(Joel Beasley at 00:41:51) No. That's a good, interesting look on it. You know, whenever I see headlines that are interesting and then I dive into it, I take a lot—if I were an AI algorithm, I take a lot of data before I'm satisfied.

(Bart at 00:42:06) Yeah. Totally. As you should.

(Joel Beasley at 00:42:08) Yeah. To wrap my head around it. And I actually don't know how much data is needed. I was hearing somebody, an evolutionary biologist talk about how much weight goes into a decision and different types of decisions. So I'm still learning about how humans make decisions, but I'm always curious about the future.

(Joel Beasley at 00:42:30) And I was thinking about quantum computing. I talked to Honeywell a couple weeks ago about—they made the fastest quantum computer. And just with new technologies or emerging things like that, how do you approach that? Do you have a team that does experiments? How do you check into emerging technologies?

(Bart at 00:42:50) Mhmm. Yeah. It's a great question. I mean, to a certain extent, I think of the role of the CTO as being the person who runs the business with the lens of technology.

(Bart at 00:43:00) You know, we have a person who runs a business with the lens of product and finance and organizational structure. But from a technology perspective, it's important to understand where is the world going? What's possible and what's not? Truly get down to the fundamentals like speed of light. What is possible and what isn't? Figuring out when these things will happen and whether or not they're relevant and whether or not you can bend the curve.

(Bart at 00:43:24) So quantum computing, I think it's awesome. I think there's some pretty amazing technology around it. The question is, is it applicable from a—just purely CTO perspective, as opposed to, like, as a technologist, oh, man, let's talk about quantum computing. As a CTO looking at it and saying, okay. Will this—is this a lever that the business can use to positively affect the experiences we're delivering to our customers?

(Bart at 00:43:40) And as a result, the value of the transaction or the value of the company. And, you know, from a quantum computing perspective, I'd say not quite there yet. From an AI perspective, for sure. Right?

(Bart at 00:44:00) I mean, if you think about the Dropbox problem, you know, people have terabytes of data, and it's like, you want to organize it. And the question is not, is the data there? The question is, what data should I be looking at now? Systems that can aggregate all the signals in your life and can start providing constructive solutions for you are awesome. The early days of this were things like Google Suggest.

(Bart at 00:44:23) Right? Where you type a couple letters and it's like, yeah, you type W. And normally, we would say Walmart, but you're in San Francisco, so we're going to say Warriors is probably what you're looking for. Getting to know that data.

(Bart at 00:44:34) Which is, by the way, why Google changed their privacy policy in 2012 to get more user data aggregated behind one user as opposed to having separate pools for search and Gmail and apps and everything. So being able to get more of that data to chew on to provide the right thing at the right time is super relevant. So from a Dropbox perspective, what we really need to do is say, hey. You just turned on your computer this morning, and you want to do something. How much information do we need to predict what you need to do to start greasing the skids to make it a better and better and better experience for you?

(Bart at 00:45:08) Right? Like, do we have access to your calendar? Do we know what meetings are coming? Do we know what, you know, like, do we have access to enough information about your daily flow that we can start saying, "Okay, here's how, you know, here's some suggestions for how you organize your day, and here's some ways that we can help"?

(Bart at 00:45:25) And so those are things where vertical AI and ML capabilities can be super useful. Things like quantum computing are still pretty far out. You know, I would say that they are becoming more and more applicable for large complex systems, like the human body or weather or things where you need to be able to process probabilistically many, you know, trillions of states and start figuring out probabilistically which ones are the most likely to happen. But at the scale and the problems we're challenged with, I don't think that, you know, I think that's still like a decade out from being viable. Now that said, you know, these things tend to move exponentially and they tend to move in fits and starts.

(Bart at 00:46:07) You know, so it's hard to say when that will become available. But you see this technology being massively applicable. So I'll give you an example. I'm looking for a Nest camera. I'm looking online. Nest cameras can now do facial recognition, and they will know when someone you know is at the door. Like, that's pretty cool. That's the aggregation of Google's deep machine learning, their optical recognition, their facial recognition, which is a fundamental part of Google Photos, being part of their platform. That platform being made available to the Nest team and the Nest team building that on top of their data streams at scale to be able to deliver a really, really useful feature to users that very few people can do.

(Bart at 00:46:52) Now anyone else can sit on top of Google's facial recognition system because it's a public API that they sell. But Google can do it way more efficiently because they get it wholesale, right? And so when you think about the ways that you can start knitting together some of these technologies, there's some really cool things you can do. But I think quantum computing is still just a little further out over the horizon to be practically applicable yet.

(Joel Beasley at 00:47:16) First of all, I very much agree on the quantum computing being too far out there because I wanted to deeply understand it versus surface level understand it. So I started taking courses on algebra and all of these different, like, linear algebra is a basis that you'll need to understand how the functions actually work, like the mathematics behind it. But I—and then my conclusion from that, like, six, eight week binge of knowledge and then talking to the experts is it's like back when there were computers the sizes of rooms. Like, it's definitely going somewhere, but the stuff right now is very machine code. We're not, like, the creatives can't get in there and do stuff easy with it yet.

(Joel Beasley at 00:47:59) We're like on day one over there. But for me, that's exciting because I get to watch and subscribe to these feeds of different college projects and what they're doing with them. And I'm always watching out for a commercialization opportunity. But let's, um, like, before we end, I wanted to continue that line of questioning. So let's pretend that, uh, I guess what I want to get at is advice from you to other CTOs on how—like, let's say there is a technology that catches their eye that does look like it could be beneficial.

(Joel Beasley at 00:48:35) What are they doing? Are they going and finding a consultant there and then recruiting them and having them do a test project? Are they getting two of their trusted engineers to do a mashup? How do you tactically, as a leader of the organization, approach exploring a new technology that you believe has some legs?

(Bart at 00:48:55) Mhmm. Yeah, it's a great question. And I don't think there's any one answer. So I'll just give you my answer, you know, how I go about it, which maybe is helpful.

(Bart at 00:49:03) I tend to think of it both from a top down and the bottom up. From the bottom up, as a pure technologist, I just like to tinker. You know, I like to give myself a project and go work on something and learn about technologies along the way. It's how I grow. It makes me happy. If I just get, you know, thirty minutes of tinker time a day, just—it's like my meditation. It's great. And so I learn new technologies that way. And so I'm always exploring in kind of a similar way to you.

(Bart at 00:49:30) Like, I'm always exploring these technologies to, like, how can I apply it? How can I learn about it? And then from the top down, I'm thinking, what is the business need? Like, where are the significant levers in the organization? And I'm asking myself, can this technology move an existing lever or will this technology expose a new lever for us?

(Bart at 00:49:49) Right. And so usually, for me, the top down, bottom up results in either me going off and getting some hands-on time, which is a little self-indulgent. Like, I can do these things on the side, but if I put myself in the critical path for anything like that in the business, I'm probably doing the business a disservice. So, you know, we have an organization inside Dropbox where we have technical team members we call scouts. And the scouts, we say, "Hey, go explore this space and come back with your findings." And it's a little bit for them like getting to scratch the itch that they have, you know. And very often, they will suggest things. So, like, "Hey, I want to go explore this."

(Bart at 00:50:28) And sometimes you have to constrain them. Sometimes you're encouraging them. Sometimes you're constraining them a little bit because there are things that are so far out there that you're like, "I just—there's no way that's actually going to ladder up to value for the business." But that said, in that top down, bottom up process, what you want is a steady stream of those things bubbling up. Like, "Hey, you know, I mean, this is how we did ASMR at Dropbox. This is how we did a bunch of our technologies." Like, people went out and explored it, and they modeled it. And they said, "Hey, here's how we think this might work."

(Bart at 00:50:58) And, you know, for every idea, you've got to take it with a grain of salt. Most of them fail. The devil's in the details. You know, again, it's one thing to do it once on a prototype on someone's laptop. It's another thing to try to do it just internally for our internal users. It's a completely different world, multiple orders of magnitude up to try to actually deploy stuff like that. And typically, what happens is as you try to scale it up, you discover new and new and new blockers because no one's greased the skids for you to do this. And so you have to ask yourself, how far out on that new frontier do you want to get into to actually try to make something like this work? So there's really that kind of top down lens, which I believe the CTO has to bring to the conversation. And then there is the kind of bottoms-up enablement.

(Bart at 00:51:43) Like, finding folks in your team who are great explorers. I've got this guy in my team who for his, uh, when he's on vacation, he will go disassemble libraries and see how they work and understand—and he'll come back and be like, "Hey, I found this thing that we can do." This is the guy who, in the early days—remember when you first use Dropbox? I don't know if you use it on a Mac.

(Joel Beasley at 00:52:02) Yeah.

(Bart at 00:52:02) Dropbox would put a little icon next to the file saying we've synced it for you. That was a private, unapproved Apple API that a very, very creative Dropbox engineer went and he punched a hole through this to add that. And it was never endorsed by Apple. They didn't really love that we were doing this, you know, but we added so much value for our users that they saw the value and they found ways to support it. And now Apple's rolling out their own kind of sync APIs and they've added that as a feature. But it came from this guy, Max, being a scout, going off, figuring it out and pioneering a thing that we were able to deploy at scale. You know? So some of these things really make it all the way through.

(Bart at 00:52:44) But many of them fail, and that's okay. They're just—you know, it's that process of kind of top down, bottom up is where the magic happens.

(Joel Beasley at 00:52:52) Yeah. That was one of the design pieces of magic because you never saw that interface before when you first saw those green check marks. It was mind-blowing.

(Bart at 00:53:00) What? Woah. That non-native.

(Joel Beasley at 00:53:03) Well, I want to be respectful of your time. For the intro, how do I pronounce your name?

(Bart at 00:53:08) Oh, hey. Uh, yeah. So that's a great question. We kind of dove right in. We didn't get to that yet. So my name is an old Sanskrit word, uh, Bharat, which means India. But most people can't pronounce that. So when I was in high school, I changed it to Bart. Everyone calls me Bart. Even the Indians call me Bart, which makes them a little uncomfortable, but it just makes life easier. So most of my life, people have called me Bart.

(Joel Beasley at 00:53:29) Boom. I'll go with Bart then. Well, thank you so much. Hey, I would, um, just because you—when you shared that story of the guy pulling out cords, that was the exact same story that Matt over at Gremlin shared that somebody at Amazon was doing.

(Bart at 00:53:48) Yeah.

(Joel Beasley at 00:53:48) So as an engineer, I would—like I don't recommend products often. That's why I feel awkward right now, but I would check it out. Like, I would have a scout or I would have somebody take a look at it. And I'm not articulating it well, but when I saw it, I was like, "This is like TDD for site reliability engineering."

(Joel Beasley at 00:54:07) It struck me as something that I got excited about and I think will be big in the future. Whether—like, I think you'd get inspiration from it, like, not become a customer, but just what they're doing was just really unique to me. It stood out to me a lot. Um, yeah.

(Bart at 00:54:23) Is it—

(Joel Beasley at 00:54:23) Yeah. I'll put it in this conversation. So if you do have a tinker time, I would check out that Gremlin thing.

(Bart at 00:54:30) Great. Thank you so much for the lead. I love to get leads like this. I'll run it down.

(Joel Beasley at 00:54:36) Yeah. I don't do it much. Only when things blow my mind, and then they come. But thank you so much. I'll add you to the list. Maybe come back next year, catch up, see how things are going. And, uh, because we had—I had pages of stuff, topics, other things we could talk about—but it was really great hanging out with you today, Bart.

(Bart at 00:54:52) Yeah. I really enjoyed it. Thank you so much for having me on the show. Enjoy the rest of your day.

(Joel Beasley at 00:54:56) You too, bud. Bye. Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn or send me an email [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.