EP 135: Aaron Levie (CEO, Box) on Enterprise AI Trends No One is Talking About Yet

28:03I think the thing that will be similar is that adopting AI versus cloud will take all the same change management. It'll take all the same meetings. It'll take all of the same privacy and policy and governance and compliance review people. It's going to be all of that. But I do notice and see one big distinction in the two waves, having kind of been at the similar timeframes for both. In the very early days of cloud, you would go into a meeting with a bank, a pharma, a government agency, and you'd say, you know, I'd love to talk to you about, you know, moving your data to the cloud. And it was sort of like, like, not, not interested, shock, like definitely not going to happen, not going to happen here.

28:48Maybe, maybe there'll be some small little workload that's like a dev test workload. Maybe some, you know, department or fringe, you know, group will use it. And so it was mostly like years and years of just like selling and selling and selling and trying to get you to understand what was so much better and all of this. Conversely, and so we're only about two and a half years into Chachabiti, let's say. So that would easily describe two and a half years into cloud if I go back to the late 2000s, let's say. If I look at AI right now, that exact same meeting with a customer or set of peer customers is like, we know this is going to change everything.

29:26Everybody's clamoring for it. Every employee needs it. The new generation coming into the workforce is asking for it. My boss is asking for it. The board is asking for it. I need to figure out a strategy. And like, it's not when, it's how. Sorry, it's not if, it's when, it's how. It's like, what can we do? I need to do more yesterday. And so totally different universes. And it is, so basically what that means is that you've bypassed all of the normal resistance to this happening that normally happens with one of these big changes. And now it's purely in the mode of like, how do we make it happen?

30:06Which functions are going to use it? What is the impact to, how do we make decisions when an AI generated the response? How much human in the loop do we need? So we've jumped all into just now that mode, five to seven years faster than we did with cloud. Now, that doesn't mean that the adoption will follow a different trajectory because, again, it's really people, it's change management, it's systems. But the reluctance and emotional kind of consternation is very different this time. And so what is the resistance when you're going into the organization? Or how would you describe beyond the enthusiasm and trying to figure out the how and the when?

30:49And like, what are the long poles in the tent? How much are on these organizations versus how much are on the commercial vendors? Yeah, I would say actually the maybe also this might abuse the analogy too much. But if I compare to cloud, cloud could do more than than what customers were asking for because they were resisting it. So you could have always implemented cloud 50 percent more than you actually were at the time or, you know, hundreds of percent more. maybe actually the reverse is true, which is I'd say most conversations now, the customer is actually like, they're jumping like three steps ahead once they see the demo.

31:26And they're like, oh, that's great. Like, we'll just like totally automate that entire thing. And you're like, okay, well, well, let's, let's, your data is not yet in the environment that it's going to need to be to do that use case. Or, you know, you know, we're still waiting on some reasoning model breakthroughs from, you know, open AI or clot or something to like, just get that extra 3 % of accuracy that you're going to need. So in this case, I think customers have pretty much jumped ahead to way closer to the end state, which is like, yeah, great. Like let's give all the employees a productivity boost.

31:58Let's figure out where we can deploy agents to the organization. Now, you know, I want to, I want to do a massive asterisk. I'm talking to the people that are talking to us, which means that like by definition, they're leaning in, they're a little bit more forward thinking. So some of, some of the anecdotes I'm sharing are from those types of customers. I think there's still an 80 % of customers that are like, whoa, like my head's exploding. Where should I start? Where should I go first? But there are a lot of customers where they're now ahead of the technology based on the set of use cases that they have.

32:29And so I wouldn't say it's like the hurdles, unfortunately, are much more like practical considerations. Like literally, like I can't just, you know, turn on an AI thing that looks at all my data and then an employee can ask the question because guess what? It's all of a sudden going to reveal, you know, secrets in the organization that they shouldn't have had access to, but they were over provisioned for access like five years ago. Or, hey, you know, AI agents are really good at doing code, you know, auto-completion, but like, am I ready to put an AI agent in production for like generating a full, you know, multi-file, you know, output?

33:09and then how do I review that when the human actually wasn't along the ride for the actual editing process? Are they going to really understand the code? Are they allowed to press submit and that goes into production? That's going to be... These are the big questions that people are dealing with. I mean, imagine how unhinged we are online where we're like, oh man, we just made a video game in three hours and it does all this stuff. And then you're an engineer in like a banking system, like, are you supposed to go implement that right now? Like, like, is that like, are we ready for like, are we ready for like your bank account to be, you know, driven by vibe coding?

33:50Like, we don't know. So, so these are, these are the kinds of things that I think every enterprise is trying to figure out like, okay, so what, what pace should I implement this stuff? You know, is my data ready? Are my systems ready? A lot of it is about, wow, the industry is moving so fast? How do I make, ironically, the industry is moving so fast. How do I make decisions right now that I'm going to be stuck with in a fast page, in a fast changing space? So a lot of it is actually like, can you get your architecture set up for like optionality in the future? Because maybe the vendor you work with today is not going to be the one that's the best in the future.

34:23You need to be able to swap out different parts. These are the kinds of conversations that we have. To that end, you cautioned, I forget if it was a blog post or a tweet or something, but that being early to AI in some ways can be a liability if you don't set up the architecture with the right abstractions in some ways. I guess as you think about the applicability of this for Box as an organization and therefore the advice you might have for others, how do you think about being flexible with this pace of change? And is it more about being model agnostic or modular in elements of the design? What would you share on that?

35:05Yeah, all of the above. So there's how we build and then how we even implement AI internally. How we build is basically we try and figure out the parts of the stack that there's probably a few different categories. And I'm making up some of the maybe the nomenclature on the fly. But like there'll be parts of the stack that just like will never change. a model breakthrough will happen and it doesn't matter. We're not going to change that particular interface with our system. So that you can just lock and load and start executing. There's other parts where you're like we think there's a best in class technology now.

35:45Maybe we're kind of like we're shaky on it. We probably prefer there to be an open source version. So that's like a spot where you probably need to be really thoughtful about how much do you lean into a particular part of the architecture. I'm thinking like your vector search, your rag kind of execution. These are spaces where there's a bunch of first movers. People started adopting them. It's not exactly clear where it all ends up in five years from now. So you kind of want to be thoughtful about that. And then there's the final part, on your modular point, which is like, you know, this space is changing every three minutes.

36:30You need to decide, do you want like maximum flexibility? So these are the model providers. Do you want to be able to have a way of, you know, swapping in different parts? There's multiple parts of Box where we use multiple model providers for the same workflow. And so you just need to then have like a high degree of optionality, a high degree of flexibility in that part of the stack. and then you have different abstraction layers between those different interfaces. But I think basically anybody in AI at this point has been building with that pattern. You go to perplexity, you change your model, you go to cursor, you change your model.

37:09I think we're all realizing like, okay, you probably want to let the user change their model or at least you as the developer have some degree of optionality there because we know how fast that space moves. And then you have to decide like which other parts of the stack to just kind of lock in on and don't worry about again. Klarna, Sebastian, who we've had on before and a good guy, great CEO, has been, I think he tweeted something recently. I think he accidentally made news in his telling of like consolidating SaaS vendors and moving more things to a gentic workflow. And I'll tell you, it set off a little bit of a firestorm internally to see if this is like a long term thing that we need to be existentially worried about investing in software because there was going to agentic workflows and systems were just going to do away with the need for the interface.

38:07And therefore, everything could be in a CRUD database in some ways. I'm curious your perspective on that as we zoom out and maybe is what they're doing, you think, broadly applicable to other organizations or is this a one of one, maybe exceptional case that they're testing something and on the early frontier? Yeah, I mean, I found it very fascinating and I enjoyed that it was happening. And I mean, I enjoy that it's still theoretically happening. I think some of his updates were that like they didn't actually do as much on the customer support side or they have some more, you know, kind of premium agent, you know, human agents that they now have.

38:46But but so I thought it was great from a provocativeness standpoint. I think we do need people testing the boundaries. I think it's the kind of thing which is it didn't cause me to rethink all of SAS because because the the, you know, I don't remember the first time I heard this, but maybe maybe it was like 15 years ago or something. I was I was meeting with an IT leader at Tesla and and they were like building. I'm going to make up 80 percent of the story, but they were building their own ERP system. And I was like, holy crap, I cannot believe that you're doing this. and maybe this is the future.

39:25And then you quickly wake up that, okay, no, this is sort of an Elon special where he needed to control end-to-end the whole thing and they needed their own... They didn't want any of the off-the-shelf stuff. And the amount of fortitude and just sheer entrepreneurialness you need to have to even want to attempt that is kind of insane. And so I think the Klarna example is more in that category. Um, uh, like, it's just like, if you go to the average company and you're like, do you want to be responsible for running your own HR system? Like the average enterprise is not going to say yes to that. And no matter how agentic, you know, something is, it does mean that you're responsible for the thing.

40:08Like, like this, the, the, the AI provider that has made the agent that wrote your, you know, your HR system is not taking responsibility for the HR system code that was generated. So that means by definition, you, the customer are now responsible for whatever the thing is that it produced. It's not going to be the AI models. You know, Claude is not, Anthropic is not taking any responsibility for how your HR system works. And if it leaks data and you get sued by the EU because there's some kind of, you know, data privacy thing, like that's not Claude, like that's Workday. Workday will do that, not Anthropic.

40:43So I think you'd have to be like, you know, which parts of software are people so unhappy about that they would do themselves or which parts of software just like are so needed to be customized or which parts of software are so unstrategic that you don't mind whipping up these like micro apps for like one off things. And so so you don't you're fine that it is built by this, you know, an agent that you can just deploy quickly and turn it off. And it's kind of, you know, it's a three month, you know, it needs to live in the world for only three months. those for the foreseeable future, next two years, let's say, that would be the only thing I think could be disrupted.

41:18But I don't think your HR system is going away. I don't think your CRM system is going away. I don't think that really changes. To your point about deflationary impact, I do think that there's a chance that you could see, you know, you could see a new crop of like, I'm going to just like, like YOLO it and build like the AI, you know, like, I'm going to build Oracle with AI agents and just like give it a shot. But that would still be a software company that emerges from that. And you're really using AI purely to just, you know, add engineering augmentation. I don't think the customer wants to be responsible for that software at the end of the day.

41:57That was my conclusion as well, which I think is good for people investing in software companies. Yeah. Now big asterisk, like in 10 years, do I want to be on the record as saying this. I'd be like, let's see. But like, but like, just, I just don't think companies want to be responsible for every problem in the world. Like, like you want, like the, the, um, uh, Jeffrey Moore came up with this idea. I think it was Jeffrey Moore that came up with it, but, but maybe he credited somebody else of core versus context. And, and like context is just like shit that like, like you just want done for you.

42:29You don't want, you don't want to worry about it. Core obviously being like, this is like existential to my business. And for most people, like, like their core business is not their HR system. And so thus you hire other people to do your HR system. And so that's why you don't really want, you know, you don't want your internal IT team being responsible for rebuilding something that you can get off the shelf and just it's done for you. That's the way I feel whenever technical founders want to reinvent sales comp. And it's just like, you're not, it's hard enough to differentiate on one vector. You know, this isn't going to be the make or break.

43:02Yeah, we, you know, We can go through all the books and talk about it. All right. I'm going to give you some quick ones and you give me your gut reaction on some of these things. Ready? Do you think the net of AI in some reasonable period of time, maybe five to 10 years, decreases the number of hours worked by your average knowledge worker? Well, how many years? Five to 10. I'm going to stick with no for now. Yeah, I think I agree. Do you think AI will flatten org structures in any meaningful way? Example, fewer middle managers, faster approval, more self-service, knowledge work. This one I could buy.

43:43I could buy. I think actually layers have gotten pretty crazy over the past couple decades. So I think yes. I think yes. And part of it is because I think that AI kind of lets you, it lets functions. My latest kind of conclusion is that it kind of lets functions sometimes do their adjacent function. And so it lets the designer write the front end code. It lets the backend developer write the front end code. It lets the copywriter generate the full white paper. So I think what that means is that you will have sort of more full stack workers almost. Like we don't really think about this outside of engineering, but I think you'll probably, I think we've gotten like maybe overly disciplined in job functions in some parts of the economy.

44:43And I think AI kind of lets you have a little bit more expansive responsibility. So then that probably means that that's a little bit of a flattening when you don't have to hop through and around as many orgs to get things done. So, but we'll see. Did you ever read the David Epstein book, Range, Why Generalists Triumph? No. It's a good one. It's a good one. Yeah, kind of. It sort of talks about, like, they give the example of Roger Federer growing up and playing multiple sports versus Tiger Woods being specialized in one. And how many breakthroughs in society have occurred from people outside of a core field or domain?

45:15leveraging the knowledge they have of that and going into another one. And so it goes through scientific breakthroughs or company foundings or all this, where when you get too verticalized in some ways, you just don't have the context of the way other people do things. And so oftentimes the breakthroughs come outside. I think work has probably gotten like a little bit to Adam Smith. And we've like, you know, we're all just doing our little part of the paperclip factory or whatever the, you know, or the needle. I think it was needle factory or whatever. Yeah, paper clip. We've all, what is it, Aaron Ross or whatever, predictable revenue.

45:52We've done that across every element of society at this point. Yeah, that's really true. Yeah, so I think probably AI relieves us a little bit of that. Like division of labor maybe got just a little bit too granular over time. And so I think maybe we get a little bit of a chance of a reset, which is like, no, I can just throw away a little bit more of a general problem at somebody and they've got the ability to kind of go and fully solve it, which, which frankly, I think is a huge, hugely exciting thing and much more fulfilling probably for, for the knowledge worker kind of class, um, uh, that, that I think has a lot of upside.

46:28What's the coolest, uh, or most mind blowing demo you've seen in the last call it three to six months where, where you were like, wow, I can't believe that's actually possible that we can do this. Yeah. Um, I'm having a, a a funny journey because like, it's like, it's like, you do, you do start to now, unfortunately, like build such high tolerance and desensitization. It's like the hedonic treadmill in some ways, right? You're just like, I, there's a, I always reference now with AI, the Louis CK bit about like airplane and wifi. And you're just like, listen, you're flying through the sky, like, and you have access to the internet.

47:05And that's the way I feel about a lot of the AI things where, you know, deep research like hallucinates a little bit. And I'm like, well, this is bullshit.

47:15It only saved me seven hours. Was that 15 hours of work you would have done otherwise prepping for this interview? Like, yes. Yeah. So it's tough because I do wish I could have my pre-Chatubiti wiring for every individual breakthrough. Because I think I would just be like, holy shit, this thing is witchcraft every day. Like, I remember, I remember like yelling at a friend who was like deep in LLMs, like, you know, a week into Chachibutti. And I was like, I was like, I'm sorry, I literally don't understand. How did it write the business strategy on a thing that's never scanned from the Internet?

47:55And like, I was just like, obviously an idiot at that point on all this stuff. And like, but I want that, I want that like, like energy again, where I'm just like surprised by everything. So, um, so I, you know, I'm, I'm, I'm like, I think from a pure utilitarian standpoint, I still love the, you know, my, you know, it's the exact example you just did, which is, which is I try and force myself as many times as possible to be like, okay, deep research exists. Deep research exists. Like, like, don't, don't do the thing where you open seven tabs and you go through all this stuff. It's just like, just send the query, go back to it.

48:32I had a, I had a thing 48 hours ago. it saved me an hour and a half. Um, and, um, and like, I, but I had to catch myself cause I was like about to just go to do my research, my normal, like, you know, I've been doing this for 20 years. You know, I know how to research things, did deep research and lo and behold, like as far as I could tell, no hallucination. I, I, I checked a couple of the answers. Fantastic. Because it's just like, boom, solve the problem for me. Like, so, you know, I'm now doing that probably five to 10 times a week on something, um, which is just an awesome productivity boost.

49:05I, I still am, I'm, I'm still probably incrementally surprised by and excited by front end design creation with AI. So, so, you know, I'll, I'll pop in a screenshot into cursor and it codes it, or I'll pop in a prompt into, you know, V zero replet. And it, it produces something that I would have like had to ping an engineer or designer about and be like, can you go work on this. And so like, like, you know, unfortunately, like my, my browser history kind of looks a little unhinged because it's like 1130pm on Saturday, I'm like on on v zero doing designs. So so that's, that's, you know, but that that still is pretty exciting, I think to me.

49:47Yeah. What do you think is more likely to be impactful in the enterprise for AI? multi-modal capabilities like image, text, code, voice, or the agentic workflows that we kind of talked about earlier? If you have perfectly executing agentic workflows, that's the holy grail. So the amount of the, you know, my workshop analogy is like, you know, in data science as an example, and anything obviously that you put into a structured database, you know, we've always had this ability to be like, okay, I'm going to go compute some large set of things. Like go run the analysis on this customer base in this region to do a thing.

50:32And like we used to, you know, we still do, but we would talk about like, okay, I have a job running and I'm going to, you know, I'll check back in in an hour and figure out what the answer was. Like imagine in all forms of unstructured knowledge work, a world where you just have jobs running, where you're just like, I just sent these 10 agents to go and like figure out and synthesize all of this clinical drug trial research to give me insights. Like you're, it's like, we were only wired to think about sending off jobs and compute tasks for like, like 3 % of all corporate work. Now imagine if you do that for the other 97 % or at least 80 % of the 90%.

51:12And And you're just like, yeah, like I sent out a job to go research this new region I want to enter as a business. I sent out a job to go in and, you know, figure out what what all of my product feedback was and then build me a roadmap that I should go look at and decide if it's validated or right. Like, like it's just going to be crazy when you can actually do full agentic workflows, whether it's deep research or executing a task or automating a customer experience. that continues to probably be the biggest area of upside. Two years, five years, 10 years. Do you think the best model in those periods will be closed source proprietary or open source?

51:57Maybe go through each, like in two years, you say closed source or open source? I think for lack of having any imagination on this, I think that I would mostly be in the camp of It almost doesn't matter because within three months of the breakthrough, you'll have an open source version. So I think it kind of doesn't matter whether the open source version did it first. You know, wow, China did it or that OpenAI did it because I think within three months, we can expect that you'll have the open source version of the closed source breakthrough. So then almost by definition, you can almost still expect then a large amount of the traffic and the use will be open source because you know you're not trading off that much because you'll still get the open weight version of what just emerged.

52:50What's the most exciting Box AI feature application workflow that you've rolled out that you think is underappreciated by your existing user base and everyone should really be thinking more about adopting? There's two features we're most excited about. The first is we have this product called Hubs. And what Hubs does is if you go to Box and you go to any file system, you create a folder, you put files in the folder, you share that with people. You have this problem, which is, well, like folders are kind of like unintuitive for the recipient because they're like, I don't know like which, you know, which third subfolder down did you put the, you know, the sales material that I'm looking for to.

53:39to use for my customer pitch. So we created hubs, which is an overlay onto folders. So you can put as many of your files and folders into a hub. And then you can go and just ask questions of the hub. And so this solves actually a very, very nuanced, highly important thing, which was if you go to a general purpose, just AI chat window, you have to do a lot of work to guess what stuff is on the other end of this chat window. If you go to a hub, in our case, you kind of know by hub, what does it have access to? So we have a sales hub, which means by definition, that has all the sales information.

54:15We have an HR hub, which means it has all your HR information. So you go to the HR hub and you ask the HR question. And then we're basically doing RAG on your documents for all of your HR documents. And we're doing RAG on all your sales presentations. We're doing RAG on all your customer support product documentation. So that basically gives you these micro knowledge portals for every topic that you want. And so that's a game changer because, again, the user instantly knows what topic is going to be covered in this thing. So you dramatically reduce the hallucination. You dramatically increase the authoritative content that will be in the hub.

54:53So that's pretty cool. And then the final one that is just working out kind of way better than we had hoped initially, like a year and a half ago, let's say, is just AI data extraction from documents. It sounds like way too straightforward for anybody listening, but it's insanely powerful if all your job is, is like review contracts, review invoices, pull out data from resumes, standardize my financial reporting documentation. We can just now automate basically most of that. That's great. Well, Aaron, thanks for doing this. I think we covered a lot of ground, so really appreciate you being there.

55:29Thank you, Logan. Good luck on the valuations. Thank you. Yeah, it's good.