Aaron Levie on enterprise AI: Box is betting the company on AI agents transforming document workflows

chat here. Bring him in. There we go. How you doing? Boom. Hey guys, how's it going? We're doing great to finally have you. You should have been on like, you know, we've had like 50 plus guests at this point. You should have been on the first one. Why wasn't I a launch partner for this thing? Seriously.

Oh, well, we're making up for today. Okay, great. Another podcast. Awesome. Good work. Yeah. Yeah. Yeah. I mean, that was the genesis, right? The world didn't need another. You said technology needed a podcast. That was the idea for sure.

Um, I want to talk about uh the enterprise AI eval and how uh the different LLMs are interacting at Box. Is this a sustaining innovation? Is this a disruptive innovation? Is Google a rival? You're gassing them up. What's going on? Uh there were like nine questions in there.

Where do you want me to uh where do you want me to start? Wherever you want. Uh yeah, I mean we're we're pretty pumped about AI as as you can imagine.

So we we've been building this platform to help enterprises store their their their data for you know a couple decades now and uh we have hundreds of billions of files within box and what the you know the one challenge with with content you know financial documents and resumes and marketing assets and contracts is you create it you share it and then you kind of never look at it again.

Y um and so what happens is you see this this sort of you know uh curve of you know people you know content remains hot for a day or two and then and then it kind of goes into an archive state. Yep.

And AI is the first thing that lets you actually tap into all of the value of all of your data you know no matter how old it is and when it was created and who it was shared with and where it is.

So for us it's this massive breakthrough which is you can turn all of this unstructured data into you know effectively usable knowledge for an organization.

Um, and so we we we're kind of betting the whole company on this idea of in the future when you work with your content, you're going to have AI agents that go out and and read your documents for you, you know, create analysis, you know, go through all your contracts, find exactly the data you're looking for, automate workflows across any business process.

So that that's effectively why we're all in and um and so we're, you know, trying to think through what are the implications in the future of enterprise software. Yeah. Um, you know, I I'm very much on the side right now that it's a TAM expansion of software as a as opposed to kind of a TAM compression.

Um, uh, I think this lets enterprise software, you know, startups or incumbents go after much bigger markets because you're now tapping into, you know, just a completely different budget pool in in an organization. Um, but uh, but yeah, lot lots happening at the moment. What about uh training a foundation model?

Did you consider it? It seems like can you talk a little bit about value acrruel versus uh foundation model layer versus application layer? We thought about it for like six and a half minutes and um that's a long time in 2025. Exactly.

So in uh in back in you know the moment we saw chatgbt like there was like one brainstorm of like oh do we need to like fine-tune our own models and or even train our own models on on you know certain data sets and um and we basically quickly concluded like no that doesn't make any sense.

Um, and the math is really clear.

It's like if you have Google and XAI, we didn't, you know, didn't have XAI at that point, but but you had, you know, OpenAI, Anthropic, Google, Meta, at a minimum, you just don't want to be, you don't want to be fighting that war, uh, from a capital standpoint or even just a deep research talent, uh, standpoint.

So, um, uh, you know, you always want to be as a, you know, a startup or any size company, you kind of want to be on the side of the tailwinds of of the market.

Um, and the tailwinds in this case are that there's like now five or six atscale companies that are deploying hundreds of billions of dollars, you know, on computers and and and training data and talent. So, you want to be riding that wave as much as possible.

And so, we've built an architecture that lets us tap into any of the AI models, you know, whoever makes them. Great. story.

Uh, you had quoted somebody, I think it was a couple days ago, that that basically delivered a pretty good prompt about a paragraph of text describing a website that they wanted and it just sort of pops out this beautiful uh, you know, landing page and the designer said, "We are so cooked.

" That was a reaction for a lot of people last week. Maybe that was a reaction if you're Studio Gibli. I'm kidding. Uh, but but you said, "Here's how this plays out. AI makes producing amazing designs more affordable and faster.

Designers get to offload the longtail work, iterate faster, and serve more clients or projects. Customers get better results faster or cheaper, inducing more demand for great design. It's basically calling Jevans parad paradox on design. Uh you is there any like context that that you feel like is worth adding there?

Because I I think they're they're generally you see these sort of big you know exciting releases or updates to models and there's like a lot of sort of doom.

But if I was if I was in high school and I like enjoyed design, I would be calling up every single SMB and I wouldn't even what I wouldn't do is say, "Hey, will you pay for this website?

" I was I would say, "I designed this website for you already and you can have it for whatever bucks and you could just be printing in like seconds, right? " So, I had a positive takeaway. I think you did too, but I was curious if you had anything else.

I mean that that that that took took it way way further than the initial thought on on you know how you how you kind of growth hack that uh which is which is I think a great idea.

I think we we you know everybody got really excited about Jevans paradox because of of you know the the efficiency of AI models the theory that as they get more efficient consumption goes up but there's an equal amount of of Jevans paradox of all of the applied use cases of AI and and so that's that gets you closer to to then the the human labor side which is which is if you just think about like most markets on the planet um I would argue uh are are sort smaller than they should be.

If the cost of delivering the service was was cheaper, you would have Jevans paradox for almost every domain in the world. Like, and let's take a very easy one. If health care was cheaper, we'd probably consume it more.

Um, like my biggest hurdle to seeing a doctor is is just it's a nightmare to to just like even get, you know, you know, a date on the calendar to to see anybody.

And so so just just think about all of the things if you could increase the productivity of that particular category um which is an which is an inverse or a correlary to lowering the cost of delivering that thing or making it more available or making it more accessible would the consumption of that category go up and I think there's not a lot of markets that are kind of at this perfect saturation level of of what the potential demand is um and we saw this in the very you know I think in modern tech we saw this let's say in Uber as an example um there was this really funny article that that some economist wrote like 15 years ago about Uber that was like the best case scenario is this takes like half the taxi revenue and it's like a $5 billion business or something and and like what the guy like obviously totally missed the forest from the trees is if you if you just make Uber you know and liquidity of of of a taxi service 100 times more efficient you know the market actually goes up by let's say you know 100x and and he couldn't kind of quite process that but we we're doing the same thing over and over again in AI where we're sort of thinking like, oh wow, you know, we made this thing way more efficient.

That means it's going to compress the size of that market.

When actually what what it will mean is is that is to your exact point, all of the people on the planet that have really bad design right now, they'll have some some entrepreneurial some, you know, high ingenuity person going and saying, "Okay, now I can actually do better design for the first time ever because it's affordable to do great design for this small bakery that that would never have hired a designer.

they never would have, you know, thought even to make their their website look good, but now it's actually affordable to do so. And so I think that's actually going to play out, you know, many many times over uh in AI.

There'll be some categories where where this where we we are at sort of equilibrium of of of sort of supply and demand, but I'd argue most categories are are this is not the case.

you know even as an example taking like another knowledge worksp space let's say you know there's a lot of AI startups doing legal legal work um you know AI for lawyers and you know one theory is okay wow is it a really bad time to be a lawyer um on the other hand if you make it much more accessible to now ask a a question that's a legal question I'm still going to follow up and say okay can you paper this now as a contract or or can you just give me a little bit extra advice but the we've lowered the barrier to now getting access to even thinking about what is the legal implication for a particular thing that then grows the demand of that space and and um you know even in the category of let's say internal legal work at a company if I could make it so we could review contracts faster with customers that we're going back and forth on.

I'm not going to have fewer lawyers in the company. We're going to actually just have higher throughput of our deal review process.

So there's a lot of these areas where where where you know you're just going to actually see an increase in demand of the overall category as AI comes and drives productivity in these uh these spaces. Talk about just hiring and and headcount expansion generally. You've have thousands of employees already at Box.

When when you're talking with the team about doing headcount planning, are you pushing people to say, "Do we actually need this person? " or if you just figured out how to use sort of these tools better, could we get away with without sort of adding that that incremental person?

Because there's been a lot of, you know, the CLA CEO came out and was basically like we're not we're never hiring anyone again and like obviously that's really good marketing, but like what what's the reality and and how are you approaching it? Yeah.

So, so we we are uh we're 100% committed to being uh an AI first company just across the entire business. So we want every employee using AI to be as productive as possible, you know, with all the right asterisks of of, you know, if you're doing production code, you're still responsible for what the AI produces.

It's not the AI agent. So So all all the all the normal, you know, kind of T's and C's on that. Um but for the most part we look at it through the lens of if we can cause efficiency in in a particular area of the business.

We want to reinvest those gains back into a another area in the business might be even the same category or same function but we want to use those dollars that we're freeing up to actually reinvest into the areas that were previously constrained.

And this is sort of, you know, even in the case of things like customer success where you're getting kind of like the first line of defense, you know, hey, I need to reset a password, those types of things that that we can now begin to automate with AI.

The dollars, at least in our case, that we that we save on on that kind of of um uh you know, you know, kind of human labor that that will actually go back into the same exact organization, but now for more proactive customer success managers that were always constrained by.

We can't we can't hire enough people to go out and do a proactive outreach to our customers and help them strategically because we have to have people that that are responding to to tickets.

And what's interesting in this case and it's I think it's an optimistic you know sort of story about AI in most cases it will be the same person that that sort of moves from one type of work to the other.

You know our our classic kind of customer success manager type roles oftentimes started in support at either our company or some other one. So, so what really what it lets you do is is recalibrate your talent to increasingly more strategic ways of of using it.

You know, the the sort of, you know, go and and um translate this marketing asset into 10 languages is a much less strategic use of time than help go work on the next marketing campaign that we want to go deliver. So, it's it really just lets you readjust into more and more strategic things.

that I think is a is an analog for what we can expect for most AI impact in in an enterprise context. Can you talk about going public? Is it as bad as people say? You went public young, you're a young guy, your hair's gray. Is it responsible of the is it because of the SEC?

Uh it's it's actually not because of the SEC, it's because of the hedge funds. Um Okay. So the that I I my my you know there's always this conversation of you know if we if we magically change the listing requirements or something of a company, you know, would that be better?

I I actually don't think I think that's a red herring. Um uh I I think that the the scrutiny that the SEC requires, the governance that you have to have, I actually think is all net positive on companies. I think that that like just lets you run a more stable, mature, you know, thoughtful business.

Um uh and and you know, any reduction in that probably if anything just causes adverse selection for the kind of companies that eventually go public.

Um, uh, I think I think the the the and this, you know, Brad Gersonner kind of kicked this conversation off a couple days ago or yesterday online of of like I, you know, about the supply of IPOs. I think this is firmly actually in Brad's hands.

Uh this is a private this is a private capital market dynamic which is as long as you have you know him and the cottos and the you know sequoas and everybody else doing very large late stage deals uh it's a you know as long as you have capital in the private market to stay private longer uh that is the the real thing that will reduce IPOs and I don't necessarily know that there's any any kind of grand economic loss for that but this is this is now this SEC has nothing to do with this this is firmly you know in the private private sector we could we could decide do we want more IPOs or not.

Uh last uh question did h how did you react to the XAIX merger? You have more followers on this platform than probably anyone I follow but Elon himself. Uh yeah, I'm curious what your take is. You've been you've been here since day one. I I I you know was was early on the uh the the Twitter wave.

Um uh I no I mean I mean I I the the uh Elon clearly likes to keep buying X. Um so it's uh he will find another way. I mean the twice now. Yeah. I have uh I I don't have a particularly strong opinion other than um you know I want I do wonder if like Tesla is the ultimate acquirer of XAI.

But but I you know I don't know how how he's going to kind of um uh you know make make all these worlds come together but um you know we'll be interesting to see and um and but no no impact on my life. Yeah. Yeah. As long as we can post. As long as we can post, we're good. Just don't turn off the posting.

Don't turn off the posting. And and and what I don't want is I don't really want like a write this tweet button. I I hope that's the one thing they won't do. I think I think that like a lot of these platforms like LinkedIn has this, you know, like help AI write your post and like I just think you get slop.

Like it would be nice if if this could be like a slop free environment or slop on X. What are you? No way. Unfathomable. No, we want handcrafted, artal, organic, farmtotable posts. Yes, exactly. We need artisal content. We need artisal content. I agree with that. Well, thanks for you.

We'd love to have you back when you have Fox News. We'll talk to you soon. Bye. Uh, who you got coming in? We got Sam Lesson. Oh. Oh, we can keep it on on X and XAI. Let's bring him in. Get a hot take. Get him out of here. the takes himself. I want to hear it. I want to hear it. You know it's going to be good.

Is that a yes or a no? He's here. He's