Uncapped #3 | Aaron Levie from Box

0:00All right, Aaron, thanks for doing this. Really appreciate you making time. Know you're really busy and it's really good to get to sit down and talk to you. Good to be here. I'm actually always here and so thank you for coming. My pleasure. So I want to start by talking about something that you've been super vocal about online lately. AI agents, AI software in the enterprise. Obviously, everybody's talking about it, but you've been talking about it in like great detail and how it relates to box. You've got this. I think you were going to say that we were talking about it first. Okay. You got the, that's right too.

0:26But can you tell me sort of briefly just like where are you sort of, where are you thinking about, you know, AI as it relates to box? How do you think agents are playing out right now? Obviously, that's what you're talking a ton about, but like what's sort of the kind of high level about what's in your mind right now? So for us, the reason I'm just so pumped about AI right now is we've been building this platform for nearly two decades to help enterprises store, manage, share, collaborate on their most important data. And, you know, that data is their financial documents, their contracts, their marketing assets, their employee records, all that data.

1:00And inside of that information contains an incredible amount of value, but most of it is underutilized. And so, you know, you just think about what you mostly do with your files. You, you know, you create one, you share it, you collaborate, and then it goes somewhere and you never see it again. And that, that we, we just see that with the life cycle of data, data, you know, starts out hot for the first couple of hours, couple of days, week or two, and then it sort of, you go, it goes into a place where you just manage it. It's actually really important to keep around because you might need to pull it up like five years later or there might be some legal reason why you need it.

1:27But, but you never tap into it in the meantime. You don't ask it questions, you can't pull out value from it. And yet it actually contains an incredible amount of wealth of value for an organization because it might have an insight that would lead to your next product discovery. It might work, you know, have information that would make a sales rep better at selling their product. It might have information that a new employee onboarding into a company will ramp up faster, but you have not been able to tap into it. So for us, AI is just this massive breakthrough, which is we can finally actually open up the value of all of this data that organizations are sitting on.

2:00We kind of are in the right place, right time in terms of having built out a platform that 115 ,000 customers trust. We're in about 67 % of the Fortune 500. And companies have been using box to be able to manage that data, automate workflows around it, secure that data. And now AI kind of plugs in right at the core of everything we're doing. So we're starting running the company and imagining the company as if, you know, if we had started the company in 2025, what would we be building? How would we be running it? What would our business model be? And we're asking ourselves like, like, are we doing everything as if we were starting from scratch in this new era of AI as opposed to, you know, the traditional challenge with being either an incumbent or some company that's been around for a while as you, you take too long to address a new technology, you don't sort of pivot hard enough.

2:47You don't reimagine your business model for that new era that's emerging. And so we're trying to, you know, learn all those lessons, maybe to some extent to prevent anything bad from happening. But then I think just generally I'm just way more excited about the upside of what's going on. You've been talking about agents. And so it's like, you've got all this data, you know, obviously with box, you know, you're storing all of the documents and the files and you know a ton about the company. So like what in your mind are the agents then like doing? Like let's say these agents get to a place that we all think they're going to get to and one, two, three years.

3:16Like what does that world look like now where I've got like box with all this data and then I've got agents that can do whatever I want them to do, whatever they kind of want to do. Like what does that look like? Yeah. So we imagine within the box environment that you'll create agents and we already have actually the something some of the core primitives available to customers in our AI studio. But we imagine a world where you'll have AI agents that you create or that are automatically created that are really, really good at content oriented workflows. So you know, what are those in an organization?

3:45Well, that's a legal assistant reviewing a contract for clauses that, you know, you don't want to agree to. That's a procurement assistant that is reviewing invoices or payment terms and then automating, you know, some part of that process. That's a marketing assistant that can look at digital assets, you know, pull out the data from those digital assets and then automate a, you know, marketing campaign sort of workflow process. So you will have, you know, millions and millions of those kinds of agents that get created. They'll help you automate the work that you do with content in every field, every industry, in every segment of enterprise and public sector, private sector and so on.

4:24That's what we're going to build out. And we think that, you know, creates this, you know, huge productivity boost for organizations because now as an employee, I can have agents sort of running in the background, executing tasks for me around my data. I can say, hey, you know, I'm a financial analyst. I want an AI agent to go do a deep research on 20 financial documents of the latest, you know, notetite took or other people took on it on a new earning cycle. And I want it to look at all that data and run a full report on what are the trends happening in the semiconductor industry, you know, where their GPU shortages, you know, what should I invest in?

4:58And that agent is going to go out, look at your data, pull back, run a report, maybe connect to outside systems, you know, the, you know, recently, opening eye has a sort of agent SDK where you can have tool use so you could pull from the web. So that might combine into that agent, you know, generating a full report for financial analysis. And again, you could imagine the same thing in legal or life sciences or healthcare and so on. So that's what the box agents will do. Then I think you kind of like, you know, this requires you to kind of extend out, you know, certainly beyond any given SaaS product or existing platform.

5:31And you say, okay, but box is not going to be the only place where work happens, you know, very clearly work happens in Salesforce and work happens in service now and work happens in Slack and work happens in in work day. And and you know, 500 other technologies. And so we're going to eventually need agents between these platforms to talk to each other. So so how do you have a an agent that can go off and pull data from five or 10 different systems to construct a full picture of of, you know, the report you're trying to run or the workflow you're trying to automate or the decision you're trying to make.

6:04And that's where the industry is so unbelievably early. But we're starting to see this emergence of either protocols or at least, you know, concepts of how agents will talk to each other where I'll go to something like a chat should be tea and I'll say, you know, run a report on this customer and it's going into box and pulling data. It's going into Salesforce and pulling data. It's going into, you know, Zoom info and and in looking at, you know, kind of being a proprietary, but but semi public data and then it runs a full report on that. And we imagine a world where you might go to, you know, three or five or 10 different sort of horizontal systems for that query or you might go to Salesforce and via something like agent force, you might run that and then that's going to go call out to box and or you might be running an ITSM workflow and service now and that's going to go call out to box.

6:47But but no matter what our agents are going to eventually have to talk to each other. It seems like anybody who's in any like enterprise software company would want these agents because they're going to be so valuable. But my guess is that the thinking would be whoever owns the data and whoever has all the integrations is going to be in the best position to sort of make the agents that do all the stuff and the company. Do you think that new startups that want to get into like building agents and stuff like that? As I'm listening to you talk, it seems to me that in most cases incumbents ought to have the structural advantage.

7:18Like unless they're asleep at the wheel basically incumbents have the data and the integration of customer relationships already. So how are you thinking about where a startup auto win? Are there places where a startup auto win other than the incumbent being asleep at the wheel? Or for the most part is it really just the incumbent should always win but big companies are tired and they're forced startups. So I think you're going to see first of all there's plenty of opportunity just in that ladder category. Like like most large companies today exist because of just incumbent to restleep at the wheel.

7:46There did not have to be white space in that Netflix found. Like that could have just been blockbuster one day saying I'm going to go digital and we're going to do the digital version of our business model. The classic of Barnes and Noble didn't have to let Amazon sort of get e -commerce and then they built out from there. So first of all, you could underwrite probably a trillion dollars of startup opportunity simply because of the kind of like asleep at the wheel kind of dynamic that's just going to exist in a whole bunch of categories. It won't be boxed because we're like we're religiously betting on this but we're very worried about competition from lots of angles but it won't be because we're asleep at the wheel.

8:22So we're going to be 100 % focused on making sure we're building out a platform that works in an agentic way. But lots of companies maybe won't be the case and it's very early to kind of figure out who's in which bucket. But you can you can tell sales versus not asleep at the wheel. Service now is not asleep at the wheel and the list goes on. Okay. So that's sort of the sleep of the wheel case. Then there's the innovator's dilemma problem which is maybe it's like the sister to sleep at the wheel. Innovator's dilemma problem is, okay, I'm an incumbent company. I really like this sort of recurring revenue thing I'm getting on a seat basis.

8:57You know, it's it's a very clean business model. I can price in a certain way and this is an opportunity for a startup to come in and say, you know what? Actually, AI has has sort of flipped the model. You know, it's a consumption oriented model as opposed to your pre buying a number of seats model and you know, a great category where where that becomes pretty obvious is things like let's say customer support software, right? Where where the innovator's dilemma dynamic is I used to sell, you know, 500 seats of software to the 500, you know, kind of customer support agents. And now we have a world where, you know, AI is going to start to automate more of that.

9:28Can my business model evolve to be able to go and support that? Because that really like innovator's dilemma, it's funny that, you know, we think about that as a tech oriented issue. It's a business model oriented oriented issue. It sort of it works on every industry and it's just it just exists because the incumbent, they don't want to do something that that basically arose their core profit. Like nobody in the management team wants to have less revenue the next year. Of course. And then have, you know, Wall Street hate them and so on. And so the CEO just keeps kind of grinding it out until basically it's too late.

10:00So it's a business model dilemma, ultimately, innovators dilemma. And there will be companies that sort of face that and it's going to be in a variety of ways. They'll be consumer versions of this. They'll be enterprise versions. And then there's a third category. So I think you can underwrite a lot on a sleep at the wheel. I think you can underwrite, you know, quite a bit of innovators dilemma. And then I think probably the biggest category will be kind of net new use cases for AI. Where there's not an obvious incumbent who sort of is in that market in in traditional software where an agent is sort of now better off serving the problem.

10:31Replacing human work basically. Replacing human work. Or just doing the human work that we never got around to. And I think this is probably one of the things that we haven't been able to quantify. Which is like how much of AI are we going to use in the future? Like, you know, Goldman Sachs or, or, you know, an economist view of these kind of analyses is like, let's look at all of human labor. And then we're going to just like figure out ingredients like how much can AI replace? It's a pretty myopic approach because it just implies that like the world is like perfectly right now, you know, we have this perfectly equilibrium of supply demand for talent.

11:04And we can all pay for the talent, you know, and it's it all kind of works. But, but actually just turns out that like we have a lot of things within box that I'd like to do that we don't have people doing right now. And we might be willing to spend X amount millions of more dollars of just net new spend. In other words, it's not just a cost saving endeavor like we should be generating new work. You'll be generating new work with AI. And so there's going to be a lot of startups that will emerge in categories where there's just not an incumbent software player or even incumbent, you know, services player that will begin to automate things that are just net new spend categories.

11:39You know, we're already seeing this in a bunch of examples, but like, you know, code gen is kind of an interesting one. You know, obviously two different takes on what the future of now, you know, coding as a profession will look like. I'm more in the optimistic take, but like, let's just say if he froze right now and you just said, okay, you know, at today's, you know, moment when you add up cursor and replet and coding him and everybody's revenue, I would argue that all of that spend on AI coding is sort of net new spend in the software stack. Like, it's all just like, all we, every company and every individual just has now decided to swipe their credit card and now buy AI to augment how they work.

12:17Right. There hasn't been replaced, you're saying? No, nobody's been replacing those categories. Again, there's going to be asterisk like, like, there could be, you know, long -term kind of, you know, consequences of this and we can talk through what will that look like. But it's net new spend. There's not an incumbent that's sort of inherently being disrupted because there was no incumbent called like the code writing incumbent and we very clearly have new startups that are actually out executing the incumbents that were in adjacent categories, all a GitHub, and delivering better products. And so, so that actually, I think you can also underwrite to hundreds of companies that are going to be worth, you know, billions or tens of billions of dollars.

12:51I mean, my mental model there is there are some activities where it's just the more the better. So, writing code, you know, selling products, like those things are the more the better. There are some that are not like that though, you know, like customer supports an example that's not really like that. Like, once all the tickets have been answered, the tickets have been answered. Yep. Agents are good at answering tickets. So, I do think we will probably have some areas quickly where like, not as many jobs are available on a per million of error basis on a company for a company. Yeah. And I guess the, you know, if you sort of looked at like global customer support and you probably broke it out by different categories, there's definitely some categories that AI, you know, is going to be very, very good at and it's going to, you know, the next task or, or, you know, password reset email or I can't log in, it's just going to happen through AI.

13:37But even for us, let's say, and we're, you know, we're going to be deploying more AI around these types of use cases. If I can save money, let's say in customer support or customer success, a lot of times the dollars that we save there are going back into that exact same function. It's just, it's moving upstream to now I want to do proactive customer success. Yep. We have always been constrained on the number of customer success managers that we can afford. Like, I would, you know, anybody in SAS knows this, you have all these like sort of ratioed roles, like how many SDRs do you have per sales rep, how many CSMs do you have per customer, and we've always, I've never been happy about the ratio.

14:14I've never been like, oh, like we have, we have too many customer success managers and it's always been because of the cost of, of it in the business model. So if I can get more efficient in one area of that function, I can just move now to human labor in other parts. Now, you got to, you have to get trained up. But, but, you know, even today, most of our customer success managers were probably customer support reps five years ago or 10 years ago, either at Boxer or a different company. And so, so I think there's actually a journey here in a lot of jobs. I don't want to be like overly optimistic and, and totally say nothing's nothing's going to happen and there's no transition because there will be areas that are more disrupted.

14:51Like, like, you know, I think we're all kind of familiar with those. But I tend to think we have more of a myopic view about how talent sort of mobilizes over time and then the new skills or the new demands that the company has as this technology rolls out. Connecting this back to the innovators dilemma business model thing, one of the things that I find at least interesting at the moment is like, you know, software can charge, you know, 10 bucks, 20 bucks, 30 bucks a month. But agents now you're comping to human labor. And so as a result, some of these agent companies can charge like surprisingly large amounts because even being a quarter of the cost of a person is like a lot more than you would have expected it to be.

15:27But you would think that that's going to sort of play out quickly over time and that, you know, we won't stay with agents staying priced relative to labor. It should eventually come back to cost. Yes. How are you thinking about this from a business model and disrupting yourself in that picture? I mean, if I had to bet on let's say, let's just make it really binary. Like, does AI remain comp that labor or does it remain comp that at sort of like infrastructure cost plus software and some margin? I'm in a bet on the latter just because competition almost sort of is going to cause that to happen.

15:58Let's say there's this human, you know, type of task that costs a hundred dollars an hour and you're just like, I'm going to build AI to do that at $50 an hour. But like, the underlying cost is one dollar, you know, an hour of compute. Somebody's just going to say, okay, I'll just do it 40 and then somebody's going to say, I'll do it at 30 and then somebody's going to do it 20 and eventually it's going to converge to, okay, we're like just going to do it like a normal software gross margin type model. So unless you have incredible proprietary access to something like, you know, for anybody who's listening, like read seven powers, it's one of the best books on this, but there's this concept of cornered resources and and if you have like a cornered resource that just like literally nobody else has on the planet, like a proprietary data set that is like everybody needs and you're the only one to get it.

16:42Sure. I think you could like always have AI agents that then get comped as people. But if you're automating, you know, work and it's code work or it's, you know, outbound sales rep work or it's, I'm going to generate, you know, marketing asset work. I think eventually it's going to converge on on more software type margin. But the benefit is that you're no longer capped by the number of people that can use a software. So the thing I'm most excited about is not necessarily that AI will get comped at the price of labor, but it's that AI doesn't have the same cap as labor. Like right now, and, you know, in the SaaS world, when you're selling seeds of software, you can only sell the number of people in the company.

17:17Like that's like that's the tam. So, so if you have a 20 person company, you're selling 20 seats. But like now with AI, that 20 person company could have like 10 AI lawyers and 10 AI ISD chargers and 10 AI marketers and all of a sudden now you're going to be spending way more on software than you were in a prior generation of that company. And so that's the, that's like the big tam expansion and I haven't seen great math yet on what that will look like, but I'm betting that that would be a very large number. I would say generally like the consensus AI view right now is like it's slightly overvalued in the short term and things are, you know, things people are too excited, prices that people are paying for these companies are too high.

17:53But in the long term, it's all good and maybe we're like, that's your fault though. Yeah, that's probably my fault. No, I'm very early. I mean, if you're naming your podcast on capped, like you are the problem. Yeah, you gave me good feedback. I might need to adjust that. Maybe just like medium sized cap. Medium cap. Medium cap. Exactly. Yeah. But like do you think that do you see us there or do you see it somehow different? Like are we, is this the internet in 1999 and it's a little bit of a bubble, but it's all going to be good in the long term or does it actually not even feel like it's over, like the excitement is too great right now because it's kind of all, you know, like these recent YC batches, everybody's working on it.

18:27It seems right, but you know, it's a fun concept of like can things be overvalued in the near term that are undervalued over the long run because by definition, the value should price in what the long run value is. So, so like I'm a hard time with that particular framing as much as just I'd say that there'll be a lot of companies that don't work and a lot of companies that do and the companies that do today's prices will look, will actually look small and those that don't, they'll look big and in a market where you have a lot of energy and hype and excitement, you'll have probably more on average companies that don't work that will get valuations that don't make sense.

19:00That's kind of true of every environment, but it's like, it's almost like there's nothing you could do about that, like because like the good assets are still going to be the good assets and in a lot of cases, you don't know the good asset until, until you see three or five attempts at a particular kind of product area. So, I, you know, it's like, it's like, yes, but then what do you do about any of that? Yeah, sure. So, okay, so on that point, so you're 20 years now in the box, which is amazing. And I think you launched 2005. Yeah. And that was like before AWS, I think slightly. We were like three months before AWS or five months before AWS and it's sad because that actually meant that we had to get good at infrastructure.

19:39And then, and then you had this sort of path dependency where like because we were going to infrastructure, we kept building out infrastructure. So it actually took us like a decade to like, like, like actually be like, no, like we should probably be a cloud company. And then, and then another five years to actually make the transition. But yes, but so you kind of saw, I mean, but you were at the beginning of cloud and you saw a cloud. And you know, then GCP comes out and Azure, I mean, the timing like looking back on some level, it's like shockingly like you were kind of first at this layer. Thank you.

20:06In cloud. So congratulations. That's really good. As you kind of look back, did it play out the way you expected? If you aside from that sort of like, you know, you wish you could have launched a year later or something like that? Are there other things knowing what you know now looking back at that last cloud cycle that you would do differently? Or are there things that you like got lucky with that were super right that you're like, think, God, we did it that way. If I have any regrets or or things that I would post more to them, it would usually just be situations where like I wish we moved faster on X project or X thing that that eventually did work.

20:37There were some key decisions that we made early on. This is more of the positive where even in even the very early days, you know, for obvious reasons, customers would say, Hey, could we run you on prem? And we we were just very stubborn and steadfast. We said no, we're like, we're a cloud company. It's all multi -tenant. It's all sass. You can't deploy this on prem. And for a number of years, it was like, is this the right decision? We didn't really question it that much, but the market certainly did and and customers were, you know, we lost a lot of customers in the process. You know, I'm extremely thankful for some very religious architecture decisions that we made early on that we stuck with because now it's set us up where, you know, you know, apropos AI, every AI capability we add to the platform, it instantly works if a customer turns it on for the entirety of our customer base.

21:19You don't have to be on like version, you know, 19 of box. Like it's just like boxes box. Everybody's on the same exact version. And when you have AI, it just like plugs into everything you're doing when you when you turn on the capabilities. So, so we've been able to kind of keep this architecture kind of clarity the whole the whole journey. Like even when we make acquisitions, we're very again, like rigorous on like we don't keep other things running with different architectures. Like it's sort of has to plug into the platform. It's got to be on the common file system. It's got to be on our cloud.

21:51And it contributes to all of the other kind of multi -tenant capabilities we have.

21:59So, off of just some some fortune architecture decisions, you know, a number of years ago. Do you think like the sort of end state of the market where you had, you know, the hyper scalers and then you have you and maybe there's like one or two others that kind of play at this sort of slightly more neutral layer across the hyper do you think the market was always destined to be all agopolis at the hyper scalers with, you know, their version and then there's going to be, you know, a player like box that is going to sort of sit neutrally across or do you think that the market could have been shaped other ways and it was just like happenstance that it played out the way that it did.