How Dropbox went all-remote and all-in on AI

Today, I’m talking with Dropbox CEO Drew Houston. Drew co-founded Dropbox way back in 2007, and he’s among the last of the founder-CEOs of that era still standing. Seventeen years is a long time to be with one company, and you’ll hear us talk a lot about all the change he’s seen in the industry.

Dropbox seems like a simple idea: By now, cloud storage is something pretty much all of us rely on in some way. But it wasn’t always that way. You’ll hear Drew talk about the early days, when Apple tried to acquire Dropbox and Steve Jobs pretty derisively told Drew and his co-founder that Dropbox was a “feature, not a product.”

Since then, a lot of companies have tried to build Dropbox-like features into their products: Google Drive, Microsoft OneDrive, and Apple’s own iCloud Drive all exist. But Dropbox has managed to fend them all off — something Drew attributes to working well across platforms instead of trying to lock you into one company.

But as Drew looked to the future, he decided to make a big bet on AI and turn Dropbox’s platform-agnostic approach into something that helps people stay organized across devices and browsers. There’s a lot of AI hype out there, and it’s clear Drew is a big AI optimist — but he has a refreshingly sober take on what AI is right now as compared to what it can or may be in a decade, and on how winding the path from “today” to “some nebulous future” actually is. Right now, Drew points out, the value from AI is definitely mostly going to Nvidia… but he’s pretty sure that’s going to change, and that the rest of us are going to come along for the ride.

You’ll also hear us talk about how Dropbox — an entire company devoted to products that make our remote work life easier — reinvented its entire conception of virtual work after covid. The company is almost entirely remote now, and Drew and I talked a lot about how he made that decision and the benefits and drawbacks that he’s seen from it.

Drew’s seen a lot in the past 17 years, and he’s pretty clear-eyed about the industry. I think you’re going to like this one.

Okay: Dropbox CEO Drew Houston. Here we go.

This transcript has been lightly edited for length and clarity.

Drew Houston, you’re the co-founder and CEO of Dropbox. Welcome to Decoder.

I’ve got to tell the audience: This is a very special episode, because Drew is in the studio with me.

We’re looking at each other. Usually, Decoder is done remotely because people are all over the place and time zones are weird, but you and I are together, which means there’s going to be a lot of cross-talking about AI. You can just feel it coming.

I feel like people are fairly familiar with Dropbox. It’s a file syncing service. It’s been around for a long time. You’ve been the CEO the whole time, which is unusual. That’s quite a journey. I want to talk about it all, but give us your view of what Dropbox is now.

In a lot of ways, Dropbox is solving the 2024 version of the problem I started with back in 2007. When I started Dropbox, I started because I kept forgetting my thumb drive, emailing myself files, all the things we used to have to do, but that was really under this bigger problem I had of: “My stuff is everywhere; I can’t find it.” In the beginning, the solution to that looked like, “Oh, I should sync my files to the cloud and across all these different devices with different operating systems that don’t talk to each other.” But fast-forward to today, and we have a lot of the same problems. My stuff is everywhere. I can’t find it. A lot of things are different. What used to be 100 files on my desktop is now 100 tabs in my browser or actually both. You have 100 files on your desktop and 100 tabs in your browser.

A lot of that’s for good reason because we have all these amazing new tools. But I think there’s also a sense that maybe we’ve gotten too much of a good thing, and we have a very fragmented and cluttered environment, and then this creates a bunch of problems, this kind of death by a thousand paper cuts. And some really important experiences have gone backward. So, take something like search. Search was actually better 20 years ago than it was today when it comes to searching your stuff for your company’s stuff because you just search your hard drive, maybe your email. But that was kind of it. But now, bizarrely, we live in this world where it’s easier to search all of human knowledge with a Google search than your own stuff or your company’s knowledge. When you go to work, you have 10 search boxes, not one.

We’re thinking a lot about these problems. How do we evolve Dropbox beyond file syncing to organize all your cloud stuff? There’s a lot we’re doing with the core Dropbox app to do that. And we’ve also been introducing new products like Dropbox Dash, which is AI-powered universal search. It works not just with your files, but it’ll search your Google Docs, your email, your Slack, your Salesforce — all of your different apps have one search bar that searches everything. And then, we’re doing a lot with AI and really personalizing AI to be able to answer a lot of the questions that ChatGPT can’t. If you think about questions like “When does my lease expire?” or “Where’s the slide from last year’s product launch where we talked about that?” ChatGPT can’t answer these questions because it’s not connected to your stuff.

But that’s what we do at Dropbox. We’ve always been platform-agnostic and trusted, and we can build a lot of that capability because Dropbox is already connected to your most important information. In a lot of ways, we’re back to the fundamentals of, in 2024, 2027, or 2030, what’s the best way to search, organize, and share content in a world that’s fragmented, and how do we fill in these missing pieces like giving you search back, helping you organize your stuff, making it easier to share things in a mixed format? If you have a Google Doc and a 10-gig 4K video, there’s not really an Airtable and iOS linked to a website or something. There’s not really a common container that holds all of that. That’s a bit of a snapshot of the fundamental problems we’re working on just around how you organize, share, and secure your content.

There’s a lot in there. I actually want to hold on to this idea of a universal container for one second. I am kind of fascinated by people’s relationship to information and how it’s stored. One of my favorite stories we did years ago now was about college students who didn’t know how file systems work, and their STEM professors had to stop and not actually teach astronomy for a day and teach them how the file system on a Windows computer works so they could use the radio telescope, which was generating files and putting them in a file system.

People have a lot of feelings about that, but that’s related to a generation of people who grew up with totally abstracted file systems on iPhones and iPads, on Chromebooks, on whatever. And you don’t even think about that anymore. You have this core feature, which is syncing files across multiple platforms in the cloud. Now, you’re saying I need to expand that to all kinds of things. I think about those things, and there’s nothing about that that says that is the same file system or information structure at all. The stuff in my Slack — I don’t think Slack is generating files in any way, shape, or form. Google Docs is a file, but it’s totally abstracted, versus a photo from my phone, which is a file but is equally abstracted. How do you think about synthesizing all of that?

Well, it’s going back to first principles, just recognizing it doesn’t have to be like this. Sometimes you can look at history, and I don’t mean history, history — I’m just like, it actually always wasn’t like this. In the physical world, you would go to sleep, and you’d wake back up, and the papers on your desk were still there. With your computer, when you reboot your computer, the files are still there. But when you’re in the browser realm, once you get to 100 tabs, you either nuke the whole browser and clear everything out because you’re declaring tab bankruptcy and you’re just overwhelmed, or your operating system updates itself in a weird way where you lose everything. But I think a whole generation of people is growing up without this basic idea: hey, you should have your stuff, and it should be possible to organize your stuff. It doesn’t have to be mayhem.

And so, there are these problems in plain sight like that. Search is one example, but organizing is another. When you think about the web world, it just evolved without really a container concept and just bizarre because files have folders, songs have playlists, links have… There’s not really an answer to that. I think we’ll look back on that and be like, “That was really weird.” And then you think about — in addition to just the fundamental container — how you have a universal container. There’s also intelligence when you use Netflix, Spotify, or YouTube — even when you sign up as a new user. The system’s smart. It knows what the world likes to watch or listen to, and then it presents you with good options. But then, when you go to work, there’s really no intelligence in that system. You can’t even wrap your arms around where is your stuff, because it lives in 10 different places. In the consumer realm, the system’s always learning from you. You’re not even filing things away. It’s just curating your experience. You just watch stuff.

Do you think that’s good or bad? I mean that question sincerely. Dropbox started as: You will have a file system in the cloud that we express natively on whatever device you’re on. It’ll look like a folder in macOS. It’ll look like a folder in Windows. We’re just extending this concept to the cloud, and that’s really powerful.

Now, we’re at this place where people don’t even know the file systems exist. The most productive people that I can think of are in control of their files and folders. They think about it. They have naming conventions for all the files they might need to make whatever document they need to produce in the end. Then, there’s the class of people that are just like, “It’s chaos, and I search Google Drive, and I can’t tell if we did a good thing, but abstracting away the file system.”

What was great about the file system, in the beginning, was that it was where all your stuff was. It was like a single source of truth. Everything worked out of the file system. And so, we’ve created a number by solving one set of problems. Files have limitations — you can’t real-time collaborate. We use things like Google Docs because you get all these real-time collaboration, and the browser brings a lot of benefits. But then, we fractured the whole stack because the web world and filed world don’t talk to each other. As we were just covering, the web world doesn’t really have persistence or organization. Any way you look at it, I think about: Alright, I opened my laptop in 2027, and I want to get to my work stuff. What do I see? Hopefully, we can do better than literally one finder window open that literally hasn’t changed since 1984 and then a browser with 100 tabs sprayed across the top to the point where you can’t even see text or titles anymore.

I think we can do better from a UX standpoint. I think these things do get better. There are a couple examples I think about. One is TV. The way TV evolved, first, it was broadcast over the air, maybe 10 channels. That was great. Then, you get basic cable and then you’re like 30 channels. But then, somewhere along the way, from 100 to 1,000 channels, it sort of lost the plot. You’re like, I just want to watch the Super Bowl, but why am I paging through all this music for channels at a time? What is going on? And Comcast just thought they were giving people what they wanted. We were asking for more channels, more stuff.

But it turned out what we needed wasn’t the next 1,000 channels. What we really needed was something like Netflix or these smart rethought systems designed to scale up to millions of things in the catalog and give you access to all of them, but the user interface uses primarily machine learning to identify what you want and recommend things and gets rid of a lot of the constraints of linear TV. So, I think we’re in that 1,000 productivity tools era when we really need a rethought system and a rethought environment. There’s a lot to learn from those kinds of evolutions. TV is one example. You could say the same thing about music. MP3s, you used to play them out of folders, and then Apple’s like, “Hang on, this is dumb. We should just have a music catalog and iTunes. It should be by artists and album and everything.” And then Spotify’s like, “Hang on, why should I have to buy these songs one at a time? I should just have an infinity everything catalog.”

And then, there were new generations after that. It’s like, well, anything I listen to or watch should all be together. There’s a whole continuum of intelligent experiences ranging from: You can still use Spotify as an iTunes-style catalog and manually curate everything. You can go all the way to the other end of the spectrum and be like, “AI DJ, just press play.” But then, a lot of stuff in between, like, oh, this artist radio, genre radio. And so, when you look at the consumer realm, there’s just this whole rich ecosystem of different ways that you can solve these problems. Then you go to your desktop at work or your workspace, and there’s just none of that intelligence or design has made its way over. We see that as a huge opportunity for Dropbox to rethink this.

I could spend the entire rest of the show talking about how much better my relationship with music was when I was individually editing ID3 tags and MP3s. I will not do that, but rest assured, I could.

That’s a big vision in the future. Dropbox today, the core business, is still selling people cloud storage. How is that split?

Dropbox has about 18 million subscribers, two and a half billion in revenue, and almost a billion in free cash flow. It is mostly people using Dropbox at work. About 80 percent of people who use Dropbox use it for work or for a mixed personal and work use case. We certainly have a lot of folks who have used Dropbox over the years to share photos or for their personal lives, but we really thought the problems that weren’t solving themselves were in the work universe. A lot of the sort of paper cuts I just described are a lot more acute in the working environment than at home.

We’re about 2,600 people. We’re 90 percent remote. We have this Virtual First working model that has been awesome for us. We compete with folks like Microsoft and Google. And with all the office suites, usually, you can get OneDrive or you can get Google Drive, but a lot of why people use Dropbox today is because it’s platform-agnostic. A lot of our customers do a lot of external sharing with clients, or if you run a marketing agency or something, you’re interacting with dozens of different organizations. Dropbox is the best at doing that because it just works on every platform regardless of what your clients or partners are using.

For big files, video production, and the creative community, Dropbox has long been standard because we handle big file syncing and that volume of data better than anyone else. We’ve built a really great trust relationship and really focused on ease of use and security, privacy. We’re uncomplicated in that we’re not trying to advertise against your data — we’re just focused on taking care of your stuff. That’s a little bit of a snapshot of where we are today. And then, we see expanding to all your cloud stuff or all of your content as a natural extension of where we started, which is taking care of all your important files.

The thing you described where you’re competing against OneDrive or Google Drive, even I imagine, to some extent, the iCloud features, there’s a very famous story from the beginning of Dropbox where Steve Jobs tried to acquire the company. You said no. He’s like, “I’m going to kill you,” but just integrating this —

Do you want to tell that story real quickly?

Dropbox was scaling like crazy because my co-founder, Arash [Ferdowsi], and I just started the company, and — it wasn’t literally out of a dorm room but looked like one. We just had this idea; it took off, and then we were scaling virally because we had this freemium model where people would spread Dropbox. And so, we got the attention of all the big platform companies, and we were invited to come down and meet Steve. I remember Arash [Ferdowsi] and I jumped in a Zipcar from our apartment in San Francisco, and we were driving down to 1 Infinite Loop in Cupertino. And we walk in, and we’re like, “We’re here to see Steve Jobs.” And they’re like, “Have a seat.” We go into the boardroom, and Steve comes in, and he starts his pitch. He’s like, “Hey, we’re like a startup with infinite resources, and you should really throw in with us, and here’s why.”

We were having a great time building the company, and for a lot of reasons, we were just like, “We admire everything that Apple’s done. We’d love to find any way to work together, but we think it’s right for us… We’re having a great time building this company independently.” And I’m like, “I’m sure you understand.”

I don’t think he loved that response. He’s like, “You’re a feature, not a product. You don’t have distribution. You don’t have this or that. You don’t control the operating system.” I was like, “Alright, agree to disagree.” Because every pair of companies has its issues, too. Apple controls Apple stuff, but they don’t control Google stuff. They don’t control Microsoft stuff and vice versa. That dynamic is exactly why Dropbox exists in the first place: you need a way to go across all the platforms. But yeah, no, it was intense. That was like 2009. But to see Steve onstage in 2011 at WWDC launch an iCloud, calling us out by name, that was just the beginning of a whole competitive period where we’re drafted up to the big leagues and fighting with not just Apple but Microsoft, Google, Amazon, Facebook, and then a raft of other companies in our space. So you get your hazing ritual as you get bigger.

That line — “You’re a feature, not a product.” I’ve heard this issued as a criticism of maybe half of the startups I’ve covered, and mostly, it’s true. They are features, and then they get Sherlocked into an operating system or integrated into some other thing, and they disappear. They get acquired or something. That has not been true for Dropbox. Seventeen years now, and you’re still the CEO.

Why do you think you’ve managed to persist?

First, the nature of our problem is that we need this platform-agnostic way of managing all of our stuff. In the ’90s, you had the Windows / Intel monoculture, but that’s just not our reality anymore. The truth is we all have accounts on all of these things, and we’ve had to get a lot clearer and more focused about, “Where is the value we provide?” If we’re just storage, then you’re right; that might be a feature. That’s not that strategic, or it’s a commodity or whatever. We’d hear those criticisms, and often, I wouldn’t necessarily disagree. I’d be like, “Wait, but our business keeps growing, and our revenue per user keeps going up, not down.” That’s a weird commodity, for example. And then, a lot of it was just talking to our customers, and often, they would say things that tripped some wire in my head.

For example, they’re like, “Yeah, Dropbox doesn’t just keep my files in sync — it keeps my team in sync.” Or, “It’s where I go to work. It’s my workplace.” And over the years, comments like that would stick, where like, “Wait, okay, that’s super interesting.” I guess that’s maybe the thing that maybe I might’ve misunderstood or Steve misunderstood. It’s like, wait, yeah, we’re not selling storage. It’s really the sharing, or it’s the fact that we can be your digital environment. That’s what people were buying.

And then, we’re just like, oh, there’s just one problem. Dropbox was never really purpose-built for that, but what if it were? So, we decided to really focus on collaboration and lean into our advantages in terms of being platform-agnostic and trust and not advertising against your content and things like that. Fundamentally, we’ve also really cared about keeping the experience easy to use and taking care of our customers. And so, we still have a loyal base of a double-digit million number of subscribers. We have a very strong foundation we’re building on.

I should just ask this dumb question. Dropbox is fundamentally in the business of selling storage. You buy hard drives in the cloud, and then people put their files on it. Do you ever see the budget for hard drives? Does someone show up in your office and is like, “We have to buy 10 more hard drives today”? What is that like?

Oh, yeah, capacity planning. Sure. 

See, you have a fancier term for it.

Yeah, we have a phenomenal team that manages all of our technical infrastructure, including all the procurement and operation of our data centers and hardware and things like that.

Do you ever pick up the phone to [call] Western Digital or Seagate?

Say, “I need 2,000 hard drives, stat!”

Not like a trading floor, but totally. And we have to manage our whole supply and not just hard drives but computers or servers and all the networking, all the elements of the stack. So, for sure, we have to go many, many levels down to the metal, and we are in the business of storage or hard drives the same way that Apple’s in the business of glass and aluminum. Those are ingredients, but that’s not really what people are buying. People are buying “having all of my stuff in one place or being able to search it, being able to share.”

This is a refinement that I’ve had to do over the years. Initially, I was like, yeah, we sell storage, we sync your files, but then when we saw like, “Oh wait, what space do we really occupy in our customer’s minds?” Their description is often very different: you help me run my business, you help me make my album, or you help me produce my videos or basically get a livelihood. And that always sends us thinking about things differently when we better understand how people actually feel about Dropbox and what they’re really hiring us to do.

When you think about your set of competitors now, I can name a bunch of weird companies. DocuSign is probably a competitor to Dropbox in some meaningful way. Who do you think of as your biggest competitors?

It’s really the office suites because 100 percent of our customers are going to have either [Microsoft] Office or Google Workspace. So, certainly, file syncing to share is a dominant category. The biggest competitive challenge is one where there’s something bundled with each of the suites. And then second, we’re 17 years in, and so the category has matured to some extent. There’s not some new continent of people who haven’t heard of Dropbox or syncing or something. And so, on the one hand, that is true, but then you look at the problems we’ve been talking about. We’re like, yeah, the cloud world has no good solutions to these problems. It’s super fragmented. The UX is flawed in all these ways, and there’s all this opportunity to bring in a lot of the stuff that has proven to work on the consumer side into the productivity realm. Zero customers have these problems solved. There are a billion knowledge workers. It’s a huge opportunity in front of us.

Let’s do the Decoder questions, talk about how Dropbox is organized, and then talk about how you’re going after these problems.

You said 2,600 people. You’re all remote. That was a big Dropbox moment. How is the company structured?

During the pandemic or before the pandemic, we were kind of a classic tech company: huge office in San Francisco, mostly Bay Area.

Like a really fancy office.

There’s a rumor that your cafeteria had a Michelin star.

He should have. That was always something we had hoped to do. But yeah, we had this awesome space, this awesome food, just an incredibly vibrant in-person community. Then covid happens. I think that sort of turned everybody upside down. First, we were preoccupied with the challenge of all of that and this trauma that everybody’s going through. But then, as we thought about it, we’re like, “Hey, wait, this the nature of…” Well, first of all, we all just went remote overnight, and this forced migration to this forced beta test of hybrid distributed work. It worked a lot better than we thought. People loved not commuting and the flexibility to live anywhere. For the first time since the concept first arose in probably the ’50s or ’60s, you could actually work from anywhere.

The stars had aligned in a really terrible way. But one of the silver linings was we can actually decouple work from our physical environment, which can have a lot of benefits. And so, we’re like, hey, the floorboards have been ripped up, this is really chaotic, but we don’t have to put them down the same way. And so, it’s a huge opportunity to really be able to design the work experience in ways that our parents never could. So, we thought about this, and then, as a practical matter, people were like, “Can I move? What are we going to do after lockdown? Can I move away? Can I not? Are we going to have to commute?”

So, we thought about it like, alright, well, there is no substitute for the in-person experience and meeting face-to-face. Our biological wiring isn’t going to change. That said, you can get a lot done on Zoom. How do we get the best of both worlds? How do we give people that flexibility but also keep the human part, the team cohesion, and build relationships and trust and all the things that are hard to do on Zoom? Also, how do we avoid the worst of both worlds, which is this two-, three-day hybrid compromise? The problem with that — I mean there are many problems with that, but one is you’re still on a leash to whatever the office space is. You can’t live outside commuting distance. Maybe you’re commuting less, but you’re still spending a lot of time in a car or train or whatever, which is a totally dead time.

And then, if it’s the same two days a week, then that’s great because everybody’s there, but then it’s super inefficient. Your CFO is like, “I don’t really love paying for a 70 percent vacant office space.” If it’s unsynchronized, it’s totally self-defeating because then you’re commuting to a half-empty office. What’s the point? What’s happened since is even worse. It’s like, “Well, I’m commuting to a half-empty office to literally be back in the same Zoom meeting, which I can come back to.”

This is my personal nightmare.

First, it’s just thinking about this from first principles. For me, a lot of my heroes are the tech founders that you’d imagine. But then also, a lot of my heroes are the management thinkers — Peter Drucker, [W. Edwards] Deming, Alfred Sloan, Henry Ford, people like that. I think a lot of that, Drucker would’ve looked at this as there’s a huge opportunity here to rethink this. So we thought about it. We studied all the remote-first companies. We decided not to be 100 percent remote because you lose the in-person part. We decided to be 90 percent. Then, 10 percent is really a more concentrated dose of human connections. So, don’t just do Zoom things when you’re together as you would in-office — really think about when you think about the most memorable experiences in your working career. It’s usually not some windowless conference room or a daily stand-up. It’s like when you’re off in some nice place or can zoom out a little bit.

That’s what we basically studied from all the remote-first companies. We synthesized all the findings into our own way of working called Virtual First. We open-sourced it, and we’ll tell you how it works. And it was a full send. This is October 2020. A lot of other companies were like, “We’ll tell you in three months, every three months.” I didn’t know that was an option, but we really felt like we owe it to our employees to give them clarity about their lives and help unblock these big decisions around where they live and how this is going to work.

But that also meant torching all of the stuff I talked about before where we came from this super vibrant in-person culture. That meant just permanently torching that for this speculative idea that we couldn’t even really put into practice until lockdown ended. So, that was a pretty wild decision. But fast-forward: today, it really works. People are 90 percent remote. We used to have 75 percent of our employees in the big tech hubs. Now, that’s like 50 percent. That means we’ve gotten all these awesome people in these places who never would’ve joined Dropbox before at all levels. Places like Boston, LA, and Chicago went from zero people working there to now all those places have 100 or hundreds.

Do they go into offices together, or are they all working from home?

Nope. We don’t ask people to commute to any office. We will do a lot of in-person programming of different kinds. We also allow people to self-organize, and we do have hubs where we do have some space. So, in San Francisco, we converted our offices into what we call studios. We ripped out all the individual workstations and things like that and turned all of our offices into these convening and collaborative spaces that are really designed more for that human connection piece called Studios. That said, I think utilization’s been pretty low.

I was going to ask about your CFOs. They’re still pretty mad that you have a bunch of empty office space?

There’s a lot of vacant office space in San Francisco, so I don’t think we’re alone in that. We have managed to sublet a lot of it. Mostly, we’re just following; we see our employees as our customers. We will support however they want to gather, but we’re finding that these retreats and off-sites and things like that are often a lot more effective than asking people to commute.

Do you think the big fancy office and the perks and the chef who should have gotten a Michelin star — was that just zero interest rate VC? You’ve been at it for a long time. You’re one of the few CEOs here who’s been the CEO of a company through all the stages for almost two decades, and there was a particular kind of excess in those early moments that is gone. Is that just a pure zero-interest rate phenomenon? Is that something else that’s changed?

I think it’s a zero-interest rate phenomenon combined with the pre-covid world where the reason why it was easier to get five people in an office five days a week is because they didn’t have an option. This whole flexibility thing wasn’t in the cards. I think that’s what a lot of CEOs today misunderstand. They keep mashing the go back to 2019 button, and they see it’s not working. Then they just push harder, and then you have this really toxic relationship. I don’t see that coming back. I think the market will tell us eventually, actually, if it is really the profitable way and best way to get great talent to have these to go back to 2015 or something like maybe, but I don’t think so.

That was the argument back then: “We need to have these perks in order to hire talent in these places. Google has water slides, and Facebook will buy you Picasso. If we don’t do these perks and give away the massages and the food, then no one will come work here.” Do you think that was true? I wonder if that was actually true.

It was true. I think we experienced the biggest war for talent that I’ve ever seen in my working lifetime over that period. Again, it’s like if you have two equal options you’re going to take, and then this one has all this cushy stuff, you can understand how that would resonate with people. It created a lot of problems, too. Even in that world, I get a sense of complacency or entitlement that’s problematic if you want to keep scaling your company and really compete. I think there were some hard lessons learned on that front that a lot of it is unhealthy.

But I think now people have voted with their feet that they value flexibility a lot more than snacks in the office. At home, you can set up your environment exactly how you want it and not just have snacks but your dog and something that’s totally purpose-built for you. So, I think forcing people back into the office is sort of trying to force people back into movie theaters — maybe you can do it for Top Gun once — or getting people back into malls or something. It was cool. Movie theaters were great, and malls were great for their time, but the world has moved on.

How is Dropbox actually structured now? You went to this fully virtual situation. You gathered people together, but how is the actual company organized?

Pretty conventionally. I have a GM of my core business, a GM of Dropbox Dash, which is our universal search product, and a lot of what I was talking about with organizing all your cloud stuff, all of our customer-facing functions like sales and marketing, and then all of our G&A functions. So pretty typical tech organization where you have business units that are really around product development, and then these horizontal functions are not going to market around G&A, so like finance and legal people.

But that’s a tougher metamorphosis than it sounds because of the basic scaling challenge that you have. You start by building. Everybody’s artisanally crafting a great product together, and then, as CEO, as you’re building a company, you have to do this metamorphosis from just working on a product to building a machine that repeatedly builds products in a super competitive environment. And so, that’s the journey we’ve been on over the last several years — really getting to that second act. There are parts where we’ve done a lot of stuff well. There are parts where we’ve had a ton of challenges, but with things like AI and a lot of the new doors that are open now, it’s super exciting, period.

This is the Decoder question: You’ve made a lot of decisions. You’re making some decisions about what to invest in now. How do you make decisions? What’s your framework?

Carefully and in a pretty structured way, actually. We had to think about this a lot, especially as we moved to Virtual First and during lockdown and so on. When we studied a lot of the remote-first companies, one thing was very clear: it is not a good idea to just take what you’re doing in person and try to photocopy that or just try to do the same thing on Zoom. Otherwise, you’re just in meetings all the time, and you just have to operate differently. And then, one quote from [Jeff Bezos] that stuck with me was that when someone asked him what’s the single best decision you’ve ever made in your career at Amazon, his response was, “The best decision I ever made was banning PowerPoint in my company and shifting to this narrative-based memo culture.”

I think that’s super important, both in general and especially in a distributed and remote world. Well, why is that? That seems very specific. What’s the difference? There are a lot of reasons. So, a PowerPoint is not as much about the ideas as much as presentation — it’s a very slow method of communicating. People can read a lot faster than they can talk. We’ve all been in those endless deck meetings where someone asks a question and that’s addressed two slides later if we can just get there. And then, Amazon pioneered this model of writing six-page memos. It’s kind of awkward when you start. They actually start the first 20 minutes of every meeting being completely silent and study hauling a document. But then, there are enormous benefits of that, which are, one, writing forces a clarity of thought that a PowerPoint doesn’t because you can smear a few bullets on a PowerPoint and be done. But to really crystallize your thoughts is a very challenging process.

It’s tough for the writer, but it makes it much better for the reader, and there are a lot more readers than writers. So you know that after that 20 minutes or half hour or whatever, everybody’s completely on the same page. They all have the same high-def picture of what is the problem we’re focused on, what is the decision we’re making. The teams had to already pre-do a lot of the work to get alignment that results in that narrative. Just in 10 different ways, you get much higher quality and faster decisions through a narrative culture or through this practice of memos and writing and documenting. When we studied the remote-first companies, most of them were like, “You have to document a lot more to be efficient.”

All the meetings start with 20 minutes of silence for you to read the six-page memos?

Yeah. Amazon was really smart about how they thought about scale more broadly. So memos are one example, but then how you structure your senior team — if you do this memo thing, you can actually have a larger executive team because you don’t have as many people talking over each other. There’s a great book called Working Backwards by two former Amazon execs, Bill [Carr] and Colin [Bryar], who actually worked with them to deploy a lot of these Amazon practices within Dropbox. And we customize them a bit. You can’t totally just photocopy stuff into your company, but we adapted a lot of it, and we’ve made it work really well. I think it’s helped us be a lot more efficient. It’s not a panacea. I think it’s really hard. It’s really challenging for a lot of people to build that muscle and write well. It takes a lot of time. So that’s often a struggle. You can lose the plot where you over-index on the document more than the decision you’re making. So, you have to put some guardrails in to constrain how much it becomes about document writing versus just moving. But in aggregate, it’s been a hugely helpful process for us.

I do like it when a company’s culture trends into memo warfare. There’s a particular kind of passive aggressiveness that you can see express itself, which I find very entertaining. I used to work at AOL — that’s all I’m saying. That’s the only thing I’ll say about that. But give me an example of how you turn that into decision-making because I ask this question a lot. Amazon comes up a lot. They have the famous type-one, type-two decisions about one-way doors and two-way doors. But the connection between, “Okay, I’ve taken that, I’ve taken some of this culture,” and “This is how I make decisions,” I find often to be quite fuzzy. Walk me through how you actually make decisions.

I think we’re the first company I can think of that really put forward a one-way door and strong perspective on what life’s going to be like after lockdown. I can walk you through how that happened. We really formalized the Amazon memo culture in the last couple of years, but that wasn’t the first time we used documents for decision-making. We had a lot of discussions about potential futures or what’s life going to be like after covid. And there was a period where we were all brainstorming and trying to figure out different things and figure out a working model. I can still remember we all were on Zoom. I was in a room above my garage in a summer place in New Hampshire in this shed. I just remember all the tiles of faces as we’re making this. Do we just torch our entire company culture and just pray that this works?

Did you find in that moment that you were spending more time with more people, but you felt more alone? Maybe that was a very personal experience to me, but I was like, “I’m talking to more people than I talk to all the time. We’re all on the grid all the time. Thirty people are always here now, but I’m more lonely than I’ve been in forever.”

That goes back to our wiring, the sense of relief we all got when lockdown ended and you could meet with people again. Again, remote is great, and the flexibility is awesome, but there’s no substitute for that. If you want to build relationships with people, build trust, which is super important. You can sustain relationships on Zoom; you can’t build relationships on Zoom — or most people can’t. So, it is a very isolating and fatiguing experience. I think it’s a lot more than just memo culture. One or two things: There’s this whole set of practices that we do to make Virtual First effective. Part of it is the in-person gathering. Second is ways to curtail meeting load. When people are in different time zones, you could end up with meetings all morning through night. So, we have this core collaboration hours concept where we try to bracket meetings within a common four-hour period.

There are a lot of things you have to do like that to really think about the work week and how you optimize it for people. There are different modes of work. Face-to-face is one mode. It’s good for certain things. There’s Zoom — you can get a lot done on Zoom. Your memos or narratives are another mode of working. Then, there’s Slack and the whole communication mayhem — that’s another mode of working. They all have different strengths and weaknesses, so it’s how you match the tasks to the right mode and design these things, and I guess the rabbit hole goes pretty deep on what you have to do to operate effectively in a distributed world.

But with that Virtual First decision, we framed the options in a document, and it was like, “Alright, we could do this Virtual First thing. We could go back to the office. We could do the ‘two, three days a week’ thing.” I mean, honestly, the discussion took 10 or 15 minutes for probably the most consequential decision that we’ve ever made, that at least I could remember making in terms of its impact on people, the company’s culture, and everybody’s lives. But it was just very clear none of these other options are going to be tenable. Going back to the office isn’t going to work. This “two days a week” thing isn’t going to work. So this is really the only thing that’s left.

We went for it and never looked back. I think we also got lucky. I mean, sometimes you make a big one-way door decision, and you go through the wrong way at the wrong door. But it’s really worked for us. Our employee retention is way up. Engagement is way up. We don’t have any problems with people commuting. We can get great talent from anywhere, as I was saying. The flexibility is like a trump card. Our offer separates are way higher. There are a lot of other factors with the market and everything else, but it has been phenomenally successful for us.

You made a hard decision about almost a year ago. You were growing. You’ve got a lot of happy employees. You maybe grew too fast. You laid off about 500 people, 16 percent of the company, and you had this quote — and it will lead us into talking about AI. You said part of it is because the era of AI is here. Explain the decision to cut down — because a lot of tech companies, in particular, have been cutting down in a way that they haven’t in a long time — and explain why AI was a catalyst for it.

First, it was horrible and not what any CEO, not what anyone wants. There were a few factors. Investments that made sense at the time kept growing or eventually got out of position with the returns we were going to see. And so, either we had built departments that were a little too big for what we needed or things like that. But I think something that really forced it was the shift to AI where it’s like, yeah, we need to invest a lot more in things like Dropbox Dash, and that means we need a lot more AI engineers and just people of a different mix.

Actually, that was the toughest part. We needed to hire new people, but we couldn’t fit it, and we couldn’t make the math work with the way that we were structured. So we had to make a really tough decision to let go of a lot of people and then make room for the investments in AI and [Dropbox] Dash and all the stuff that we wanted to ultimately make the company successful. But it’s brutal.

That restructuring, there’s a bunch of stuff we’re doing. We can see the future is coming. That’s right after what you might call the ChatGPT moment. Light bulbs went off for everyone. Walk me through that part of the decision. You’re like, “Oh, this happened. We see what’s coming. I’ve got to flip the company.” Is that a bunch of memos? Is that a long conversation?

It was one long memo that I wrote, and that was the end of a thought process. But I feel lucky. I was born into the PC era. I was in middle school and high school as the internet came along. Mobile and cloud made Dropbox possible because the second you have a smartphone is the first time you have this issue of having two devices like a laptop and a smartphone. How do you deal with data? AWS launched the year before Dropbox was founded, so our board was right at the center of two massive waves. There are not a lot of good things about getting older, but I’m 41 now, so I remember when I was 24, I was starting the company and the title wave kind of feeling. And I’m like, it’s happening again.

It’s very obvious that things are different, and it’s going to shake everything up, make some things a lot more challenging, and open up all kinds of new pathways for every company. Folks that adapt quickly become the new winners in the new era. Dropbox, in a lot of ways, should never have existed if you think about it. We didn’t own an operating system. We just had every weakness you could imagine, yet somehow, here we are. And those are the most consequential periods in tech because every 10 years or so, the ground unfreezes, and big changes, big sweeping changes, are possible. And then, things kind of hardened up again after a few years.

And so, it was very clear both from my lived experience and then also that I love studying the history of tech and business and all kinds of things. But it brought back to mind things like Andy Grove’s Only the Paranoid Survive talking about how Intel navigated their transition. They started out as a memory business, and then they found that there was a strategic inflection point where the world had changed, and they basically had these competitors that were faster, better, and cheaper than they were, putting them in a really tough spot. And so, they decided to move to microprocessors. Andy goes through the whole story of that difficulty, that transition, which sort of sounds good in hindsight because you’re like, yeah, they lived happily ever after. But that’s like Google getting out of search. Intel getting out of memory is like Google… just a crazy decision.

So this is something. In those periods, those strategic inflection points that Andy talks about are the most important moments for our industry and in a tech company’s life. So, I had some experience going through that and having to surf those kinds of waves. And so, it was very obvious after using ChatGPT, like, whoa, okay, we’ve uncorked machine intelligence. Let’s go. And so, the mechanism I used was writing a 3,500-page memo talking about how we need to shift Dropbox to being an AI-first company. We need to launch Dash. ChatGPT and things like that are amazing, but there are a lot of questions it can’t answer, as I was saying, because it’s not personalized to you. There’s a big void that Dropbox can fill by building personalized AI. So we got after it super quickly. And then, one of the advantages of being founder and CEO and controlling shareholders, you can pivot the company pretty quickly, and that ended up being really important.

I hear the comparison to, “This is mobile and social all over again.” Half of it, to me, feels like it’s super real. You can see it. You can see a bunch of new startups, companies, and ideas, and application models are moving from one place to another. And then, part of me says, “This is wishful thinking, and everyone just wants to get away from mobile phones because they’re controlled by two dominant players.” And boy, it would be cool if we could do something other than exist at the whims of the two dominant players.

What’s that mix in your head? Is it 50-50? Is it 80-20?

It’s all of the above. That’s what makes these things so hard to navigate: AI will be the best; AI will be the worst — just like the printing press did great things, but you can write really bad things as a result.

I felt the same way about crypto. Everyone wanted it, and I was just like, “This is stupid.”

I didn’t. I was a crypto bear.

But you don’t feel that way about this?

No. To me, it’s clear that AI will be bigger than any of the transitions of PC, cloud, mobile, or the internet. It’s more in the fire or electricity or industrial revolution-type category. But again, this is why it’s really important for founders to study because when you’re starting your first company, everything is new to you. You think everything’s different, but then you zoom out with a historical perspective. You’re like, oh, wait, no, these are very obvious recurrent patterns, and there are life cycles of companies. I think you can learn a lot about companies from studying how empires rise and fall. There’s actually a lot of the same dynamics, like complacency, entitlement, lack of hustle.

You get very big, and then you do fail.

And we were living that honestly. But coming back to AI, it is the real deal. We now have a human brain and a silicon brain that are very complementary. All kinds of new opportunities for automation, all the benefits that we imagine are going to happen. And I think a lot of the challenges that we foresee or that challenges we’ve had in recent eras — making sure these technologies are a force for good on balance — we’re going to live that again. But you live that with every technology like cars. Cars are great. They give all kinds of mobility. They also are one of the leading causes of death in the world. But it’s an awkward navigation. But we’re going to go through that process.

These previous transitions — cars: you could just go build a car, and many companies have tried, many companies have failed, but they could start and produce a car. Mobile: you could just build an app; you could download Xcode; you build an app; you can play it. But for AI, there are a whole bunch of dependencies where to start an AI company. You actually almost end up starting an OpenAI or Google or someone who owns the H100s in a data center or Azure or something. That dependency feels different if you’re trying to compete with the AI companies. I will get in trouble for saying this, but so many AI startups are like ChatGPT raptors.

You have a dependency on some of these models. Do you think, okay, part of the curve here is breaking away from those dependencies, or that will be the infrastructure that everybody uses?

I think this is also why it’s super important for founders to understand business, and a lot of the curriculum you learn in business school is actually very relevant. I never went to business school, so I can’t really speak to that, but I think it’s a huge question of: where is the value going to accrue?

Right now, it’s to Nvidia.

That’s right. Because they have the scarce resource, and I think it’s also going to accrue to the people who have the relationship with the customer. So, the application part of the stack. And then—

Wait, no — I think that’s a good question for you. That feels less certain. It’s certain that a lot of the value will accrue to Nvidia. They make the chip that’s in hottest demand. The customer’s paying money for the applications yet to be seen. I don’t know that that is a definite in this whole mix.

It’s interesting because you have sort of these simultaneous positive and negative indicators. So ChatGPT speed ran from 0 to 100 million monthlies and probably a billion in revenue faster than any product, probably faster than the iPhone. And yet, it’s also flattened out a little bit. And then, yes, the amount of investment that has gone into AI versus the amount of incremental engagement or revenue is totally upside down. The way I think about that is, I mean, the internet is actually a pretty comparable thing, which is as soon as the internet came along, people’s imaginations went wild. Even though the first experience was very primitive — this was even before the web — it was like maybe you could send an email. That in and of itself was pretty cool and useful. I send a postcard around the world in 100 milliseconds — this is unbelievable.

But then, a lot of farsighted people were like, “Guys, this is going to change everything. We’re going to get all our news on our computer. We’re going to be able to watch any movie or music ever created. We’re going to be able to have groceries delivered to our door.” And then, all those things turned out to be right, but they happened a lot later — the time constants were wrong. So, I think keeping that in mind is really important. And then, specifically along the way to Spotify, there had to be a Napster and then Kazaa and Limewire and Pandora, where you had some false starts as a lot of the other stuff got ready. Netflix started out in DVD mailing because the broadband penetration and the enabling conditions and distribution weren’t there yet. And so, they had to wait for that to catch up.

With self-driving cars, a similar thing happened. After it sort of became generally agreed that self-driving was going to be a thing, people were like, “Oh my God, in 10 minutes, no one’s going to ever drive a car again. Everybody’s going to be out of work.” And, to this day, the best self-driving thing most of us have used — very few of us got here in a self-driving car, or maybe have even been in a self-driving car, but all of us have Google Maps. Google Maps has probably had a bigger impact on driving than self-driving cars because these things kind of follow. People think that Level 5 autonomy or a fully autonomous car is step one when it’s actually step 100. Only 10, 12 years after all this is this even starting to become real.

I think we’re going to see the same dynamics with AI. There’s this, again, these MBA academic things. There are a lot of different frameworks out there where these hype cycles, that trough of disillusionment because we had this meteoric explosion and high expectations since the starting gun went off of ChatGPT, yet that hasn’t really translated to a lot of new products or revenue other than ChatGPT. So, there’s going to be this big crash of disappointment. You’re already seeing AI companies get off the rails and implode, but these are very familiar rhythms and patterns in these tech cycles. We’re going to see all of that.

There are some very familiar criticisms. Dropbox has even faced a few of them. I have a Dropbox account. I put my data into your system. I have some amount of trust that Dropbox isn’t going to do anything with it. AI is all about taking action on the data. So, Dash searches. I think you have some video editing features that are AI-powered. Those necessarily involve other people’s models and data going in different places and being manipulated. That feels like a big difference in kind, and I don’t think the average consumer, particularly enterprise user, has a lot of trust in those controls yet.

That’s the first thing on everyone’s mind, every customer’s mind, on our mind. There are enormous privacy, security, safety concerns with AI, and there are also enormous privacy, security, and safety concerns as we move our stuff in the cloud to begin with. And so, this was not a new phenomenon for us.

But I’ll give you an example — I think this is a very funny story in its way. The House of Representatives had to instruct its staffers this week not to use Microsoft Copilot because Microsoft is just rolling it out in the Office suites. They all use Microsoft. We’re banning this tool that is very… it seems fun to use, but the work of the House of Representatives has security clearance issues.

To me, that’s just very funny. Microsoft is like, “Here’s this tool.” And then the answer is you might be sharing classified information in a way that you should not be sharing.

Totally. And they probably were.

AI can make bullet points out of it. That problem — it seems totally foreseeable but also impossible to foresee. How are you thinking about that?

The same way that we were thinking about moving people’s stuff to the cloud to begin with. It’s super important to give people transparency and control and to really take ownership of the full stack and experience. This happens all the time, too, as we move to sending our credit card number over the internet, a lot of well-justified concern, and so we see the same thing happening with AI. So, transparency: How do these things actually work? Who’s going to have access to the data and then give people control? Do I want to disable this entirely? Do I want to allow only these AI providers? Do I not want any third-party AI providers at all?

So, versus the opposite of, oh, it’s very opaque, or companies don’t give care to these issues or blow themselves up on pretty predictable obstacles. We have a lot of experience. We’ve only dealt in these kinds of existential concerns, so I think it’s pretty familiar to us, but transparency, control, and then educating people on what’s actually happening. And then, as far as the AI stacks themselves, there are a lot of options — you can use OpenAI or a lot of public cloud options. There have been phenomenal advances in open models and open source — like the Llama models for Meta — but the whole gap between the closed-source and open-source models has been rapidly closing.

I think it’s OpenAI that you use right now. Would you transition to running some of the open-source models?

We play with everything, and we’ll give the customers a choice. So it’ll be “yes and.” Because we partner with Open AI, and to be clear, we give our customers very explicit visibility and control. Whenever you use any kind of AI feature, we only use AI for that specific purpose. We have all these protections around, and then extra agreements and things with AI providers around, like here are the rules of the road for how customer data is used. We launched Dropbox AI principles. We make a bunch of public commitments around here’s what we do and don’t do. I think these are table stakes things that are pretty important.

Dropbox has a big advantage because we have this whole technical infrastructure. We operate one of the largest cloud infrastructures in the world, cloud services in the world. We were the one car going in the other direction almost 10 years ago, and we shifted from the public cloud to our own hosted infrastructure. The reasons we did that were to be able to have more control. It is more cost-efficient. It’s a lot more performance. And then, it’s also very favorable as the AI world has come up because a lot of that stuff you do to build a giant hard drive for everyone is a lot of the same things you do to index the known universe of content and apps and is a lot of the same stuff you do to build AI.

So, large language models — like serving, fine-tuning, inference, and things like that — can build on a very similar foundation to what we already have, which is another big advantage that Dropbox has when deploying it. So, yes, we’ll absolutely use open-source models, but there are a lot of benefits to different closed-source models. Your customers may have lots of different preferences in terms of who they want to work with and how they want to handle their data. For us, it’ll be a lot of following what customers want, again, giving them transparency, control, and choice.

At the beginning of the conversation, you talked about sorting out people’s tabs and helping them understand all the different containers on the internet. That requires you to sit a different layer in the computing stack, maybe at the browser layer, maybe somewhere else. Is that something you’re actively thinking about?

Well, we’re in the browser in terms of an extension. Dash — just stepping back a little bit of what Dropbox Dash is — gives you AI-powered universal search. So, what does that mean? Well, you wire up all your apps. You connect to all the things that you want to be searchable. It gives you one search box, and it will search everything for you instead of having to go to every place manually. Then also, we’ll provide you universal answers. So, you can ask natural language questions like ChatGPT, and it’ll give you answers. But we’re also in the browser, so it’s not just a search box. For customers that want it, we will give you a start page for work.

When you hit a new tab at work in a normal browser, what does it show you? It shows you politics and news and sports and tweets, and when you’re not in work mode, that might be what you want. But when you’re working, you’re like, “I’m trying to focus on something, and this is definitely not what I want to see, so why don’t I have a smarter start page for work?” That’s what we do. We help you. We’ll give you an overview of your day. We’ll help surface the stuff you’ve been working on a lot or that needs your attention and give you a better cockpit for work. We certainly have the technical capability to build a browser, but that’s a big customer request.

There are companies chasing “What does a reimagined browser for AI look like?” And I’m just listening to you describe the vision, and it’s like, why would you want to have some big dependency on Google or Microsoft or Apple to run your extension when there’s another thing there? It’s really those dependencies that you’ve been fighting against the whole time.

You have a dependency on macOS not blocking Chromebox, right?

Yeah. At the end of the day, every company’s going to have dependencies. There’s not really an option to be fully self-contained. And at the end of the day, you share customers with these companies, and if you’re building something that millions of people want and are happy with, then it’s not in the platform provider’s interests. They can have sharp elbows. They can disenfranchise you and cause a lot of problems for sure. But we’ve always tried to have a collaborative relationship, and as long as you’re providing customer value, then it’s harder — not impossible — to do flagrantly anticompetitive things.

Do you think the regulations we’re seeing in Europe and some of the antitrust actions here will change your relationship with the big platform companies?

I think we all want a level playing field. So, yes. A lot of their recent attention, I think, is good in terms of forcing some level of interoperability and giving customers choice. I think these are all good things, or I think it’s a bad world if there are just a few platforms and they can control the entire experience and own distribution and provide their own first-party alternatives, privilege them, and tax their competitor. It’s like, choose six of those things, not all 15. I think it will be a good tipping or bending things back and do a spot that’s better for innovation, for sure, because I think it’d be very hard to start the next Dropbox today, given the environment.

I think that’s very true, and I think that’s honestly one of the reasons I’m paying attention to both the platform shift to AI and the regulatory shift. Because the opportunities to create new kinds of companies — it feels more open than it has in a long time.

My wish list is still pretty long, but it’s important that the next great set of startups can be created and it’s not a great world if all the value of the innovation is harvested by 50-year-old companies. I think that’s not a great balance.

Well, that’s a really great place to leave it. You’ve given us more time than we bargained for. Drew, thank you so much for coming on Decoder.