Why the "SaaS is dead" narrative is completely wrong - Aaron Levie [box]

0:00AI is too important for the rest of the world to say we're only going to use the API calls from three companies in the US. There's a bunch of people secretly cheering that this is exactly what you'd want to have happen to design the regulatory review process for AI model releases. Intelligence is going to be multifaceted. It's not going to be one or two models doing everything.

0:21Billions Host:Today on Billions, I'm sitting down with Aaron Levie. He is the leader who rapidly transformed Box from a pure SaaS business into a cutting-edge platform for AI agents. We dive into the high-stakes battle for sovereign AI, government regulations, and why the narrative that SaaS is dead might be completely wrong. Aaron, super excited to have you on. Super excited to have you on. I know you have a very interesting take about AI. I'm curious to get your last take on what happened with Entropic recently. Because I think you said it was a big win for open and lightweight models. So can you maybe elaborate on your thoughts on the topic?

1:10Yeah, so I think, first of all, the Fable model is kind of an unprecedented level of intelligence and capabilities. So so the clearly Anthropic, you know, designed an incredible and trained an incredible model. But the U.S. government effectively enacting an export control, which meant which meant that non U.S. citizens aren't able to use the model is really kind of a completely new precedent and moment in A.I. regulation where we've never seen that before for any kind of frontier model being held back legally from being available in the market. And the implications are, first of all, it's almost entirely impractical for any AI lab to be fully responsible for, you know, the nationality of the customers that are using it.

2:03At a minimum, it would completely obliterate the API business of Anthropic because, you know, the idea that you would be able to control the end customer, you know, through your partner ecosystem is very, very difficult. So it effectively has the impact of making the model unavailable to the world. And so the risk is if you were either a major app developer and you want to be able to make this technology available to your customers or you're a country anywhere around the world that wants to continue to ensure that your ecosystem is able to build on top of leading frontier AI. this poses an all new risk, which is at any moment, you know, can that AI model get pulled back from one government or due to one government?

2:48And we just never had that precedent before. You know, AI has been this sort of very fast moving, very market based ecosystem over the past couple of years. And so, you know, I think, first of all, the U.S. government probably very quickly needs to to improve the situation with Anthropic. I think Anthropic is quite motivated to resolve whatever the particular issues are. But even when that happens, it's easy to kind of think in a year from now or three years from now or five years from now, how many more versions of this might we see in this landscape that will play out? And so it sort of makes you wonder, well, if you're, you know, a country, Germany, France, Japan, you know, UK, and you have the capability to, you know, be it at, you know, frontier or near frontier AI from a talent and an infrastructure standpoint, do you need to be, you know, building these kinds of models yourself and make sure that you have them available to, you know, businesses and users within your country?

3:51I think it really kind of introduces that question in a much bigger way. And we've already seen, obviously, examples of this. China obviously developing, you know, kind of great open source models, Mistral, you know, doing the same. But I think this highlights the increased importance in that dynamic and the idea that you can have sovereign AI or AI that you control. Now, there's one, there's sort of one caveat to all of this, though, which is, which is, you know, you're not going to, you're not going to take that big of a sacrifice or a discount on your on the model's intelligence for that benefit.

4:27So if Frontier from US in a kind of a closed model environment stays 10, 20, 30, 50 percent better than the next best model that you could make sovereign, I think people will take that risk and they will use the Frontier model at all times. So this only works if open weights models or sovereign AI can remain close to the Frontier and much closer than it has historically. And I think we're seeing some examples with Kimi and DeepSeek and a few of these new models coming out of China and Nemotron. These models are doing a great job kind of staying at the forefront, but they're still not Fable level.

5:11And that's going to be the big question for the ecosystem over the coming kind of months and years.

5:16Billions Host:Do you feel like it's only going to be a matter of performance, like staying, let's say, as you said, 20, 30 percent behind or also a matter of timing? because what we're seeing right now is like when the first model went out, you know, like the first lab to catch up maybe took like 12 months or like a bit more than a year. But now I feel like models are catching up like within a couple of months. So do you think it's only a matter of difference or also a matter of how fast can other models catch up on the best models? Well, I think, you know, there's a big debate, you know, and if you watch, you know, Dwarkesh talk to Dario or Jensen, you know, you sort of see examples and contours of this debate, which is, is the frontier and the open and the sort of, let's say, the best in class open weights model, how close are these tracking versus are we seeing a divergence?

6:14Are they six months behind? Are they three months behind? Are they one month behind? Or how much is it really just a factor of how much compute you have for inference and kind of test time compute as an example? Because there's different ways to kind of really drive the workloads in these models to get better results. I am generally actually optimistic that the frontier open weights models will be able to stay relatively caught up. I think it's of such economic and strategic importance for countries like China, maybe individual companies like Mistral or even NVIDIA to make sure that that is sort of maintained.

6:59And we have not yet seen a breakthrough that is so either proprietary or kind of undiscovered by the rest of the ecosystem that keeps this completely closed off. You know, I think there's obviously a conversation around, you know, RSI and and what, you know, what does self-improvement look like? And does that kind of lead to a kind of an accelerating takeoff? Like whoever, you know, is able to design these systems that self-improve first, then obviously there's a compounding dynamic to that. At the same time, it's hard to imagine China not wanting to stay in the race in a very big way. Like it's unlikely that China says, you know what, we're totally fine with Anthropica and OpenAI being the only vendors of AI in the world.

7:45And, you know, we'll be closed off from this ecosystem. So at some point, you know, you can just throw enough money at this problem. You know, you can China could just say we're going to spend 50 billion dollars and just and just make sure that this is this is of such critical national security importance that we have to make sure that we're throwing compute at this problem. We're going to kind of, you know, sequester the top researchers and scientists and and make it happen. I think it's unlikely that that wouldn't work. And I haven't seen any evidence that suggests otherwise.

8:19Billions Host:so your view like when it comes to china is uh is what regarding like the the chips because for so long like uh you know there's been kind of an embargo on chips and it was not possible for them to get them and they were getting it through like singapore or like all these kind of things like what what's kind of your view because it's like it feels like eventually like uh i mean in In the US, it has always been a very pro open market. And now it feels like things are moving away from it. So I'm curious to get your take. Well, if you and I think a lot of very smart people believe this, if you if you believe that AI is is both a national security threat and opportunity.

9:01And it's a, you know, of critical strategic economic imperative that can lead some some people down the path of, well, then you actually want this to be a closed environment. You need to keep the chips. You need to keep the models. And then I do think it's of national security level of importance. But there's this interesting tension, which is if you're also not the exporter of the technology, then the ecosystem effects start to favor the non-U.S. companies. And Jensen made this case in the Dorkesh podcast about a month ago. So and actually, I probably lean more toward Jensen's side of this argument, which is, you know, if you just play this out, AI is too important for, you know, the rest of the world to say, you know, oh, yeah, we're only going to, you know, use the API calls from, you know, three companies in the US.

9:54It's just too critical of a technology. So sovereign clouds have to get built out. GVU clusters are going to get built out globally. you know countries are going to want to train their own models like that that's an inevitability so if that's an inevitability then you have a choice do you want your technology stacked do you want your architecture to be the architecture that powers all that or do you want to force the rest of the world to have to go and design their other architecture and if that happens who stands the game the most obviously china because they will just eventually stand up enough enough compute enough infrastructure and make that happen now i think this is a very um uh worthwhile debate and i am only like like 80 percent confident in my opinion uh the the 20 percent where i'm not confident is you could you could possibly squint and argue that america is just so far ahead in compute and chip design and now we're starting to kind of you know figure out how to do fabs um and you've got elon with terra fab and we we know how to you know do the orbital data centers maybe in the future that maybe there's one scenario that you play out, and that is that we do actually just have the API for intelligence and that we control the access to it.

11:08And we can make sure that only our allies are using it in the right ways and we can prevent the bad actors. I think there's certainly a scenario that looks like that. That's what many of the kind of AI safety kind of ecosystem is sort of hoping for as the outcome. And I totally grant there's some percentage possibility of that. But I kind of default to this is such critical infrastructure and so important that it will cause other countries to have to just go their own direction if the US isn't enabling that.

11:44Billions Host:Yeah, personally, I'm more on the side of people should let Jensen win, you know? He has the chip, should sell them everywhere. But what's interesting is you can already now see the downstream impact. So Jensen was the first to experience this. But, you know, Anthropic is now the next to experience it. It's the same exact thing. So if you already believe in the precedent that you shouldn't export a chip to China or some other country designing, you know, kind of leading AI, then it stands to reason you probably also aren't going to let the leading AI model that could help train the next model or help crack systems be available to those countries or, you know, certain ecosystems.

12:27systems. So we now have, you know, because of the chip export controls, you have a lot of the foundation for now what we're going to be dealing with. And again, the risk is that creates a very unstable environment for countries globally. Now, again, as long as we have the best models, you know, maybe you just, you can just, you know, power through that. But it creates a lot of incentive for other countries to want to respond in kind. Yeah.

12:52Billions Host:And do you feel like looking at entropic like marketing in the recent months i mean it's been pretty common for models to make this huge announcement that their next model shouldn't be released because it's too dangerous and we saw that you know with open ai so obviously what it does is like everyone gets excited about the model yes entropic did that uh six months ago and now they they've done it again with mythos and and fable like do you feel like uh their marketing went a bit too far and it's just like uh going against them or do you really feel yeah yeah it was right no no no i was just wondering because like uh david sachs like uh was sometimes like working at uh with the the white house like what he was saying is that essentially like they they emailed the entropic team and they told us like hey can you please uh deploy your patch on that specific model and apparently the entropic team hasn't been like that responsive.

13:48Billions Host:So we're hearing like kind of two sides of the story. And I think like in the marketing of Entropic, it's been very real from the start that they want to be seen as the good guys versus evil Samatman, which I think is not super fair when you look at the global environment. So yeah, I would love to get your take on this. Yeah, you know, this is one of the more complicated topics right now because so so let's start with the easiest part. Unquestionably, the rhetoric that has come out of of Anthropic and and other kind of safety researchers created the environment that we're now in where where if you if you kind of scare the crap out of the government by saying this is the most dangerous thing we've ever seen and has all these risks.

14:41And then the government sort of you know, you know, we can debate how kind of compelling the particular jailbreak is. But if they eventually discover that there's a jailbreak on what you claim is now the safe version and then it defaults back to the unsafe version or it can do something that it perceives as unsafe. Well, that's that's clearly a byproduct of creating the atmosphere where everybody is concerned about this. So like unquestionably, it's coming from that tone and that that kind of sort of, you know, kind of rhetoric. So that's the first layer. So like that's, I think, quite straightforward.

15:17It's pretty obvious. And so we're kind of dealing with now the consequence of that. The interesting thing that I, what I don't know, though, is, you know, is it plausible that there's a scenario here where actually Anthropoc prefers, weirdly, prefers this outcome, which is now we have a precedent that has been set that models at a certain threshold of capability do have to go through an approval process. They do have to go through a review process. They do have to be approved and vetted by the government. That's not so different from what the regulatory sort of lobbying has been sort of pushing for from many of the AI safety researchers.

16:01And so interestingly, if you kind of study the journey of this whole space, the top AI researchers in safety and kind of, you know, regulations and controls probably would have designed a system that says the government is going to approve the release of models at a certain capability level. We're going to do lots of red teaming. We're going to kind of get in a room and we're going to agree that the thing is safe to release. And nobody had kind of created the catalyst previously to just build that environment. So I think there's like one other layer, which I like hold out a little bit of percentage of my thought process on is like, maybe this is exactly what Anthropic wanted to happen.

16:44You know, we're all dealing with the day-to-day drama of, you know, wow, I can't believe they responded that way. Or I can't believe Amazon is the one that like, it's like, you know, we're caught up in the sort of all the little leaks and the messages when actually like you zoom out and you're like, this is sort of how you would design the regulatory process of AI model releases if you were a deep AI sort of safety research oriented person that really was concerned about these kinds of risks. Now, I happen to not be in that camp. So I think this is sort of a bad outcome. But I could totally be convinced very easily that there's a bunch of people secretly cheering that this is exactly you know what you'd want to have happen to design the regulatory review

17:27Billions Host:process for ai model releases yeah it makes sense and um when you look at like the the competitors like talking about i think i think for like open source models in uh in china there is always like a question regarding the data and where it goes and uh how it's used etc and uh when you look at companies like Mistral who have positioned themselves, you know, really towards more enterprise and helping enterprise like build on top of their models. Recently, we've seen, I think Cursor is actually like a really good use case of how you can specialize a model towards something very specific and get massive value on top of it.

18:11Billions Host:So how do you see this play out? And do you feel like Mistral is potentially the one that is positioned the best toward this because they've already a deep, you know, roots inside enterprise and governments, etc. to develop on top of their own data? Yeah, I really like, first of all, I really like Mistral's kind of market opportunity much more today than maybe three years ago. It's actually interesting as AI gets more important, as the capabilities get more capable, more effective, as agentic workloads cost more money, all of those things start to favor a bit more of a kind of multiplayer ecosystem as opposed to sort of just one model from one or two players.

18:58So I think Mistral is a kind of a net winner in this. I think, again, one or two of the kind of Chinese labs are, I don't know if they're financially net winners, but they're kind of net winners on using those models. There's these companies, BaseTat and Fireworks, they're net winners because you're going to go and do RL on one of these foundation models that are open and then kind of, you know, sort of post-train it for your particular use case. I think that's a winning strategy. We saw that from Cursor. we're seeing this more and more in a lot of the kind of applied AI companies. So I think the big update of the past, frankly, month to two months, and then the past four days is actually like intelligence is going to be multifaceted.

19:41It's not going to be one or two models doing everything. And there's even, you know, winners really in the applied layer, like cursor, cursor is kind of this great case study now of we're going to build a harness, It's going to get really good at that one thing, which is coding. That harness can route workloads to different models. And oh, by the way, we're going to have our own cheap model that's fast. We're also going to let you use Fable and let you use Codex and GBD55. And then I think that's kind of a pretty good stable equilibrium in all of this. And then everybody has to compete on some dimension of, OK, my model is better at this use case.

20:17My model is lower cost. My model is better for this particular sovereign requirement. Like that's a very logical sort of outcome we would end up with in this landscape.

20:28Billions Host:Yeah, I agree. And I think it's good. Like it's healthy for a business because I think a few years back we were thinking that it's going to be kind of a winner takes all, you know, where you have like one model that can beat them all. But now we're seeing, yeah. I think the past two months with like the token cost growing, the sort of, you know, benchmark improvements we've seen from open models, like, and then the Fable thing as there to this extra kind of element that we didn't really factor in here. I think these are actually all very good things for the overall AI space. And I'm actually curious because you made like, I mean, you're at the forefront of AI and I've seen you talk about it from the start and you also made the transition like very quickly from SaaS to AI and to agents at Box.

21:19Billions Host:How has your vision evolved and how do you go from a pure SaaS business model to something that leveraging AI and hence has also different cost structure for your business? Yeah, so for us, our whole strategy is agents need access to critical enterprise data to be able to do their work. They need access to your documents, your marketing assets, your research, your contracts, your financial records. And so we built a platform that helps agents get access to that data and do useful work with it. So that's kind of the overall value proposition. We do this in basically two ways. One is you can use agents that are directly built on Box.

22:06And those agents use the frontier intelligence that we've been talking about. It could be from OpenAI. It could be from Anthropic. It could be from Gemini. It could be from an open model. and we will plug into all of the agentic ecosystems that you want to use. So they can plug into Cloud Cowork and Codex and ChatGPT and Mistral's chat system. So we plug into any AI system you want to use and we have our own agents that let you do a variety of kind of document use cases and unstructured data workflows. So that's our sort of strategy. And the really exciting thing is what it does is it lets you just unleash all of these new capabilities on your enterprise data that weren't possible before.

22:49You can give our agent access to all of your research or all of your marketing assets or all of your internal policies or your entire product roadmap. And then any user can just go and either ask a question of that data or generate new content or generate a sales presentation or go and process contracts and see where the risk is. So anything you want to be able to do with all that unstructured data, our agentic platform effectively is powering for you.

23:15Billions Host:And do you feel like the way people are going to work inside companies are going to evolve towards? Because I think like where, you know, like Box is super well-structured is you have like all your documents into one place and you can like manage them. But now you're kind of building the company's brain because you have like all these files that are like, you know, like basically like accessible and queryable in real time with MCPs, et cetera, et cetera. So do you feel like companies are going to move from a place where you can really like access and structure documents? So for example, if you look at Notion, essentially a lot of people are writing their SOPs on Notion, writing like guidelines for marketing, et cetera, et cetera.

24:03Billions Host:But in the end, is that really useful when you have agents you can query directly? I don't think so. And I see like Bucks very well positioned toward a new category of just like, yeah, just business information in general. And but I'm curious to know how you perceive this. Yeah, I think I think your kind of angle there is correct. And, you know, we kind of think about it a little bit like how, you know, people work. So basically, you know, people work with, you know, some roles you work with contracts. Other roles, you work with marketing assets, other roles, you work with product documentation, other roles, you work with invoices.

24:44And we do all that work. And agents need access to that same data and that same information that we work with. And so I can't really create a separate universe just for agents to operate in because I, as an end user, need access to the same data that they're working with. And so this idea that there's a sort of a separated agent brain that it has access to and I kind of lightly contribute to, that's sort of unlikely to work. What has to happen is the agent needs to be able to access the same data that I'm updating or that my colleagues update or my customers updating, which means that you need some kind of shared content system or knowledge system or file system that the agent and the people have access to.

25:23And so what is better than just literally the file system that we're already using, but built in such a way that both agents and people can use it? And so within Box, we've basically built that platform. We, you know, primarily started out building it for people. Then we actually added applications to it. So we've already kind of built this machine user orientation over a decade and a half. And now there's just one new type of user, which is an agent user. And so the whole idea is what if you had a single file system that people, applications, and agents all leverage? And those agents could be from any external system, like a cloud coworker and codex, etc.

26:00Or it could be from, again, my own internal agents that I've designed within Box. We're kind of indifferent to that approach. We obviously kind of monetize them differently. But that's the whole vision. And so this idea of, you know, your company now has this digital brain for agents and people to all work together is what we're building. And then there have been things that we've had to like meaningfully change. So for instance, you know, a year ago, we could barely spell markdown. Six months ago, you could like click a markdown file and look at it. And now we let you edit the markdown and let you save it and let you give it access to an agent.

26:33And the next thing we're doing is making sure we've got really good HTML compatibility because you want to let the agent export a bunch of stuff that you want to go visualize. So you have to improve the system to make it easy to work with and share and collaborate with the agent. We've had to launch, you know, you know, obviously an MCP server. We've had to dramatically improve our CLI. We're playing around with new kind of file system ideas that agents can access. So there's a lot of underlying technology that's needed to make this all work. But obviously, you know incredibly exciting for us because of all these new use cases that we can go empower

27:06Billions Host:yeah no agree 100 and when it comes to the narrative you know that sass are dead uh what do you think like people get usually wrong with that narrative um i think that there's two things that they get wrong uh and and then one thing that they get right actually so let me start with the right thing so we can move to the positive the right thing is is that there is some sass that should get kind of compressed because at some point, if I ask an agent to, you know, do X task, if X task used to be the thing that that SAS did holistically, then I'm obviously going to use that that SAS system less. So that's sort of inevitable.