Dropbox's Drew Houston on Dash launch: self-serve AI search for SMBs and the 'universal context layer' strategy

number one AI agent. If you want your AI to handle customer support, go to finn. ai. Um, our next guest is Drew Hust from Dropbox. Welcome to the show. Drew, how are you doing? What's happening? How's it going, guys? Great to see you. Uh, we've been excited for this one. Yeah, very, very excited.

Um, where should we start? Are are you are you a Whimo fan? Are you an Uber fan? We were debating like where does the consumer experience go?

I mean I I I know you're probably in the position where you might not have to choose, but uh it it's an interesting question in my mind of like will people want the the you know the super app for that can get them the robot taxi or the human uh long term or if uh folks just wind up moving over to Whimo entirely?

Yeah, I mean I think it's going to be interesting to see how fast it diffuses. I mean I think there's still like that um or that that like quality problem and they're getting that like third nine, fourth nine, fifth nine. Yeah.

I mean that's really what's and Carpathy talked about that on why agents are going to be further away than people think. And so but I mean I think it's I mean SF it's like pretty well established. So I think that's probably a sign of what's to come. Yeah.

Uh, did the did the Carpathy uh did the Carpathy uh podcast like update you on Dropbox's strategy at all or was it something that you were already hearing like rumbles of and you were like, "Okay, now I have like permission to fully write that to long-term memory or how were you processing it?

" You know, because I think they were like there it there was nothing there that was like completely, oh, no one has ever given this take before. It was more like no one has ever said it in this context with this authority.

It was it was uh it was somebody whose opinion holds a lot of weight that isn't like waiting didn't feel like he's waiting out for the next like big tender offer, you know. No, totally. So, it was like it was felt like less conflicted than some of the other opinions that that get shared. Yeah. Totally. Yeah.

No, I'm very aligned with it.

I don't think there was to your point, I don't think it was like a lot of new um uh or you know, I agree with a lot of these themes that the industry's been thinking about for a while, but um yeah, sort of like a splash of sobriety on the space and and I think self-driving is a good uh comp in some ways for like why this why AI or AGI or this like fast takeoff might you know not be as quick or on on the kind of timelines that people are thinking about because yeah, you know, Carpathy talks about this March of Nines like it's the same amount of work to go from like 90% reliability.

That's like one unit of work. Then to get to 99 is like the same amount of work. Then 99. 9. I mean, we've certainly seen that in our um as we've worked with AI and as we just work on reliability in general. I think that's that's very consistent.

And I think yeah, the self-driving car took the world or capture the world's imagination like 15 years ago. And yet it's these like more intermediate forms of automation that are that come way before autonomy that create a lot of the commercial value um and get a lot more adoption.

So, for example, like there's like billions of users of Google Maps today. Uh, but you know, maybe thousands of people are going to get in a Whimo today. Um, and then same thing for like, you know, highway assist or what Tesla's done or some of these intermediate forms of self-driving that have maybe millions of users.

And so, we think about that a lot as we deploy AI. And, you know, you don't want to be either too early or too late, both with tech in general, but also with automation, for sure. Yeah. Yeah.

Yeah, it feels like if you watched the Carpathy episode on Friday and you walked into the office on Monday and like ripped up your whole plan, like something's maybe going wrong with your leadership style. Um, but but it is an interesting frame to think about it in the decade of agents framing.

And I'm wondering like you've been at this for two decades now, correct? Like roughly. Uh, and uh, and so you're you're like fully equipped to think in decades because you have literally thought in decades multiple times, right?

And so I'm wondering how you're thinking about the next decade like in 10 years what if things play out kind of to the midcase and we get a lot of really powerful AI systems but they don't come next quarter how do you set your company up to actually capitalize fully over like a nice long decade of just solid execution.

Yeah. Um, well, I think when AI first came along, people's imaginations immediately went to like, you know, sci-fi movies and sort of the one omnicient omnipotent AI assistant. Um, but for a while, yeah, I think like looking at I was surprised when I was, you know, 30 when self-driving first came out.

I was like, "Oh, yeah, maybe maybe millions of people will be off the road. Maybe we'll never see another human behind the wheel uh when we go to work. " Um, and it just like took a lot longer.

And and then I've been working with like even when you're just designing like systems with engineering and and certainly as you start to apply machine intelligence, you see these dynamics where like the demos look super good um and sort of convince you that it's like right around the corner, but then when you actually scale these things up, it's a lot harder than it looks.

And so we saw that even before AI with like Dropbox to begin with like we could make a very good demo or a video like in the first couple months but then to roll it out at like mission critical uh and massive sc to roll out such a mission critical product at massive scale um you know you can't have a bad day and that just takes a lot more engineering and and we see that with our AI products today.

So I mean we're and along the you know we want to build the omnicient you know omnipotent AI agent too but when we actually look at you like look at customers and look on their screens the more basic problem that people have that AI can help with today is just that like your screen's a mess like work is a mess you have like 100 tabs open um you know you're working across Google and Slack and your email and Dropbox and you know dozen other things and you know the question I started with is like how much is AI helping with that?

Yeah. And the answer is like zero or negative. Yeah. The the other thing is I'm sure reliability was like a core focus for Dropbox during the early days because if you're like a fi online file storage and then like 1% of the time like a file just like disappears into the ether.

It's not saved and then the the the user you know might have lost it forever and that's like going to stick in their brain forever. And so like reliability at Dropbox means like 99. 9999999999999 forever, right?

And and it feels like the industry in you know spending a decade building SAS and building SAS that was pretty reliable and and having like a bunch of infrastructure that makes it easier to deliver reliability maybe forgot how like how important it is.

And if you have an AI agent that that does an amazing job 90% of the time and just totally botches botches it 10% of the time, it's not ready for production.

And so you can have these magical and I think in the context of like customer support is a big one is like is your agent ready for production if it if like 95% of the time it it it solves the consumer's problem, but 5% of the time it makes them like pissed off and they're just like talk to a human.

talk to a human, you know, and they're just running down that that that conversation.

But yeah, well, I mean, even if once you get up to a million users, even if you have a 95 or 99% success rate, that's still like thousands or 10 thousands tens of thousands of people having like a really terrible experience like every day. And this was actually part of the motivation why I started Dropbox.

Like step one was not like create Dropbox. Step one was like use all these other services and like have all these near misses or or things were like wait um what happened to my like tax return? It was just here a second ago. Where did it go?

And going into the support forums of these other competit like well these other companies, these other products that claim to do this on paper. Um but then you go in the support forums and it's like a battlefield medical tent.

Like people are like, "Oh my god, what happened to my wedding photos and where's this and where's and I was like, I like can't believe that, you know, these things are so brittle. Um, I'm going to be putting my most important stuff in here even just as a customer of one like me. Like I need something that I can trust.

" Um, and and then that took a lot more like engineering rigor and discipline um than we'd seen in the space. We've been getting into barnyard analogies around here. I want to take you through one and uh I'd love your feedback.

So uh we we we think about a lot of the a lot of the dynamics between the various players in enterprise software and foundation models as uh potentially mimicking the experience of a fox and a hen house. And uh and it might be the job of the CEO of the hen house not to let the fox in the hen house.

And I'm wondering if that resonates with you at all in terms of uh if I'm a foundation model company, I say, "Hey, I'd love to provide a bunch of great AI tooling to my customers, your customers, both. Uh I will need all of your data forever.

Uh and I will need, you know, to significantly erode your mode or your customer retention. Like is the job of the CEO right now to protect the hen house or go on the attack and be more fox-like? like how help me walk through this from your perspective. Well, I mean I think when it comes to using AI, it depends.

I mean the first is um and it depends what your business is doing and and how you're using it and so on. I think um I mean certainly the CEO's responsibility to like strategically position their company and mitigate risk and things like that. Um but certainly every CEO is thinking about how do I like usefully deploy AI?

Um, and so I think it helps to have um an understanding of the technology. I think it it's important to like know who you're working with and what their incentives are.

To your point, you know, hen houses like um you know, I think a lot of people have apprehension about using some of these or about working with the a foundation model company whose whole business revolves around like taking data and sort of grinding it up into pellets for training the next one.

and you know you watch Sora videos or whatever and you're like that comes from somewhere. Yeah, probably not. They're artists.

Um so um so I think and you know certainly in our business that trust and privacy um and not having incentive conflicts of like not advertising, not training foundation models is super important as you know all of our customers think about you know who should we partner with.

So is there a little bit of a tension there where um you you need to keep customer trust high and let them know that uh we're going to get you these AI features but in the Dropbox way that is actually fully integrated and we're not just going to take your customer record or your data and sell it off the back of the truck uh just to pull something forward.

And that actually layers into like the customer experience. Like you might be a couple months behind on a certain thing, but it's going to wind up with a better experience over time. Is that accurate? Yeah. And I think it it's all the above. I think it has to it starts with just like your business and your incentives.

Like you have to design some of this in from the beginning. Like you know we could probably be make more money or be more profitable if we did advertising for example, right? And so and a lot of scaled, you know, the predominant scaled business model on the internet for certainly for consumers is is advertising.

And we're like, you know, we're just going to draw a hard line. We're never we're just not going to do that. Um because of this trust conflict, you know, this incentive conflict we would have and even if we're even if our intentions are good, it's like customers just have to wonder, right?

And so um and then that becomes a big advantage because then we've been like living up to that for the last 18 years and we have a track record of What what is the specific concern if if you were running effectively display ads? Is it privacy of like I don't want you to use my data to target me with more ads? Yeah.

And really this is up to our customers, not us. It doesn't really matter what we think. It really matters what they think.

And I think yeah, everybody feels a little weird when it's like um the equivalent of like, you know, even just like the TV in your living room if it starts showing you like if it's like a smart TV that showing starts showing you ads you didn't ask for. you know, it's like, whoa, there's an invasion of privacy here.

Um, and so you have to be super sensitive to the psychology um of your users. But I think at a more fundamental level, yeah, the other thing is like non-targeted. I'm imagining I'm imagining the worst possible scenario that you were 100% not doing where the the file storage company edits ads into my files.

So, I'm like, wait a minute, I saved my high school transcript and now on page six of this PDF, they edited in an ad. Well, the the other thing is like non-targeted ads are really frustrating because I I remember in college I bought a Kindle and there was an option. It was like, do you want the ad supported?

I went with the ad supported. This is a very Amazon move is to be like, okay, for $100 you can get the ad support or for 120 you can buy it. And I'm a college student. I got the bougie one. I got the no ads one.

I so but but then I was getting like the the ads that that were that when I had this that were popular were like the most it was like romantic novels and stuff like that like books that I wasn't reading and so my channel would be sitting there and it be like just absorb my entire life so you can target me with something cool for once.

Yeah. Uh interesting. And and the other thing is I think you have to focus like you have to really be selective about the problems you solve and like figure about okay how do we take what we do today and build on that in a way in a way that makes sense for like today's context.

So for us I mean we always took care of your files these are already like your most important information but as we all know like what used to be 100 files on our desktops now 100 tabs in our browser.

Um and then the big a lot of the pain points we have are like you know instead of one search box or at home I can Google anything from one place search all of human knowledge. I go to work, I've got like 10 different search boxes. My experience is totally fragmented. Um, and that's what I was living personally.

That's what my company's living. That's what all of us are living. I mean, look under my Zoom window. It's all, you know, we're all dealing with the same thing. And so, we're like, all right, let's extend AI into that new world in like a valuable way. Let's give people that one search box.

Let's and then another problem with AI is that our industry is training these is spending trillions of dollars to train these models that can do like quantum physics but don't know about you know can't find your Q2 board tech and so and the missing link there is like AI these models can only be as smart as the context that you provide them.

Um, and that's really the bottleneck and like no one's focusing on that and we so that's why that's how like we have this new product Dropbox Dash um which really aims to close that context gap and finally connect your AI to to everything at work. Uh, and so on the surface it's a AI assistant. It's a search engine.

It connects it knows about you. It knows about your work. It knows about your company. It connects to all of your work apps. Um, and unlike with a consumer AI tool or you know where there's a lot of questions, it can answer a lot of questions but a lot of questions it can't answer.

So if you ask like when does our office lease expire or where's that slide from last year's product launch chatbt can't tell you cuz it's not connected to your stuff and you know and then copilot doesn't know about your slack and Gemini doesn't know about your Salesforce and so that's where Dash comes in to build this like universal context layer.

Um, and the front end is really around search and and an AI assistant because that's what people understand. Um, but here's but we're like this is a natural evolution for us.

Dropbox has always been about organizing and sharing your information or like organizing sharing your most important information, connecting to every ecosystem, um, being trusted, having this track record of reliability and security.

Okay, how can we what what is the manifestation of AI in our world that actually makes sense and where is there a new problem that hasn't been solved?

I I think that's just like the perfect strategy for you right now because I that we had this like search was great and then we wound up generating so much data that search I personally not a judgment on you particularly but across all of my services uh whether it's iMessage Gmail we talk about it all the time uh it's not to hate on you this is a n this is a technologywide thing uh search got really slow and and not as accurate and it would there were so many cookies that it was picking up like it would just get confused.

And then we got like GPT5 deep research like the thing that can go work for 20 minutes to get you exactly what you need and it's like okay now we need to actually bring these two things together and get really performance something in the middle and then size them and you see this with the model router in the in the consumer product but to actually bring something like this in the in the business context makes so much so much sense.

So, uh, I love that the pitch is not is not ASI for your documents, you know, it's like, yeah, I mean, we're excited about that too eventually.

But, you know, if that's, you know, in self-driving, they have like level five full autonomy, like totally self-driving, human out of the loop, but a lot of the commercial value gets created at like the level one, level two, like, all right, let's give people a working search box.

Let's actually give them AI that actually knows about their work. Um these are going to be the big businesses that get built along the way to that you know super intelligence. Yep.

Zooming out uh and and kind of going away from what's happening at the product level like what is the what is your updated kind of strategy around running the business and capital allocation. I know you're the single largest shareholder at somewhere around 20%.

You've been buying back a bunch of shares but like what is your what is like the the strategy and kind of pitch on on that front? Well, mainly investing for growth. I mean, Dropbox has always been a pretty profitable business.

I mean, we took a lot of the consumer internet playbook in the beginning and built these like self-s serve viral and productled um motion that was able to scale to millions of or you know, hundreds of millions of people, millions of businesses.

We have half a million paying businesses on Dropbox, 2 and a half billion in revenue. Um, but you know, the the new Yeah, there we go. Yeah, we got to hit the actual half billion of revenue creating productled growth. We get we get excited about big numbers. The godfather. Are you the godfather of productled growth?

Is that fair to say? I don't think I'm allowed to call myself any Okay. Well, I'm allowed to call I'm calling you the godfather of productled growth. No. No. I mean, if if if you if anyone listening is not familiar with the Dropbox playbook, like please go study it. It is a clinic uh for for value creation.

It's a fantastic story. Yeah. And you know, we're doing that again.

We're running a lot of the same playbook with Dash and I talked a little about search and um and AI the sort of AI chat piece but we're also just thinking very fundamentally about like all the things we kind of lost as we went from kind of the file and PC world into the browser and the cloud and and it's not just search or AI there's even just the concept of organizing things as you were talking about people like somehow we all got convinced that searching for things um is like uh a substitute for organizing things and it's just not really true in that you know all along the way we had um fi files have folders songs have playlist but you think like links have right there's no like container like a collection of um there's no way to like have a collection of mixed format stuff if you have like a Google doc and a 10 gig 4K video and an air table um there's no way to still organize so we're we have something called stacks uh probably about to show up on this video um that does that but yeah I mean just looking at these very basic basic problems hidden in plain sight and you know you give people a search box that works let people talk to all their stuff in natural language organize their stuff for them you know go beyond documents to organize your images and videos um and so in a lot of ways I where I'm putting our capital is just building the version of Dropbox that I would want to have today which is and Dash is step one in that direction yeah and there's I mean we've covered a number of companies that have been going after enterprise search and it feels like you were and many of them have been able to achieve pretty wild valuations um in a short period of time, but of course you're not asleep at the wheel thinking I don't I don't want to I don't want to dominate that market too.

So it's it's it's awesome. Well, and I think that the big thing our customers really want or or where Dash fits in there versus Glean or some other companies is like you don't have to spend 6 months and 50 grand to get started.

Um, and so one of the things we're launching today is like a truly self-s served version of Dash where you can just download it, get your team up and running um, link your apps and just go.

Uh, and that's something I think that's really been missing because there's a lot of, you know, there's a bit of a knife fight in the enterprise um, around a lot of different things, but then we look at any most of our customers and in the SMB segment or lower midm market is a huge swath of the market has no options.

Um, and so we think it took a lot longer because it's like hard to make these things totally self-s serve. Um, but that's what we've done with Dash and and again it's sort of like it's the version of Dropbox I would build for today's context. Fantastic. Um, we won't keep you any longer. Obviously, you're very busy.

We really appreciate your congrats to the whole team on the launch. Appreciate it. We're super proud of it. And uh, next time come on in person, come hang out in the Ultradome and put on a growth clinic.

I'm sure a lot of people would benefit from uh learning some of the some of the story because uh from the Godfather himself. Yeah, it's a fantastic one. The Godfather. Uh thanks so much for coming on the show. We'll talk have a great We'll talk to you soon. Up next, we have Adam Fry from Open AI. That rhymes.

I like that. The product lead for Chatbt Atlas. We're diving into the browser wars. uh the the uh the new web browser from uh OpenAI uh had a phenomenal response. I don't know if you