Box's Aaron Levie on Predictions for the Next Wave of AI: Will Foundation Models Be…

happy to provide your evening entertainment.

We are clearly in this this window that is going to be relatively temporary. I don’t know if it’ll be two years. I don’t know if it’ll be five years. I don’t know if we if it’s already over. But, basically, you have these windows more or less once a decade. We had it in the the PC boom, basically the eighties. You had it in the web boom in the nineties. You had it in the mobile and cloud booms in the kind of mid two thousands and two thousand tens.

And in each one of these moments, there’s an architecture shift that happens in tech that creates really the only window of opportunity you have for new kind of platform scale, you know, large franchise kind of companies to emerge because you need a change in the technology industry for new insurgents to be able to enter or else, basically, the incumbents will just naturally gobble up all over the market.

So we are in at one of these moments with AI, which is a period where we are going to see not only breakthrough technology, but the breakthrough application of those technologies That is as much going to be an incumbent’s game as a startups game this time around because the incumbents have a lot of the data and a lot of the workflows. So it’s even more competitive, I think, than the prior periods of these windows.

But you will have this moment of of opportunity where startups will emerge that will be able to figure out a set of use cases or a way that that they they should deliver functionality on AI that an incumbent will not be paying attention to, will not may maybe go after the in these windows where you can build now very large companies. And so we’re in one of these windows, and it’s sort of these are the only times when when you really have that level of of clear opportunity in front of you.

And so now I I was most of this is a call to, you’re gonna be working nonstop if you’re in one of these companies. Like, there is no chance that that you should be on sort of anything other than just pure survival and execution as a as an organization.

I would say that there will be some foundation model companies, but not nearly the magnitude of of the application layer companies. And much of that is is due to the trends we’ve already seen, which is, you know, the moment you have people like Zuckerberg that are are literally willing to spend billions and billions of dollars commoditizing the model layer, it it becomes very hard to sort of figure out, well, how do you how do you differentiate in that space enough where you won’t, you know, kind of be taken out by by one large training run from an OpenAI or a Google or or a Zuck.

There will be, like, niche or industry specific, you know, sort of approaches you could take or some very, you know, kind of specific domains you could go after. Like, I think there could be categories where, you know, maybe the big incumbents are more nervous about going after audio because there’s obviously going to be lots of of interesting conversations around copyright and whatnot. But I think for, like, the the pure play horizontal LLMs, I think those will largely be subsumed by the the the big players with maybe room for one, two, three independent companies at scale that that are not in the in the hyperscalers, but there will not be room for 50 companies.

That that’s just not gonna be possible.

So we are building an architecture that lets our customers do what you just said. So effectively switch between models that they wanna use for different purposes. I don’t think you’re gonna have complete commoditization of sort of the personality of the models or the the sort of style and the response to the point where then, you know, if you’re a user, any given response could come from Gemini versus g p d four versus, you know, Claude. I I think that’s probably less practical. You’re gonna be more wired into a particular model for some some use case as a user of software.

What what we did basically is as as soon as sort of this wave started eighteen months ago, we basically started working on an AI platform layer that that connects the data in Box securely with any AI model, starting with OpenAI’s models. And then o over time, we will be opening that up to other AI models as well. So if you’re a customer and you say, okay. I really, you know, find GPT four is very good at legal, you know, answers, but Gemini is very good at pulling up metadata from documents.

Those would two different use cases you would have with AI within Box, and we would give you the ability to kind of interact with multiple AI models to go do that on your on your content.

You know, interestingly, I’m not necessarily seeing that. In fact, in fact, I’d probably say the opposite. For the first time ever, we’re now talking to customers that have been cloud holdouts that are say that that previous to AI, they they said, I’m fine with this data being on prem. I feel like safer in inside of an on prem environment. AI is kind of the final death knell to to that particular approach. If your data is not in the cloud or in a cloud ready form, I think it’s gonna be extremely hard for you to get the full value of AI on your information.

We have our own internal benchmarking that we do. So we compare models using a set of business documents, but the rate of improvement has been truly incredible, just in the past eighteen months. One critical metric for us is basically, token window size. We we rely heavily on how much data can you give the model and can you get back from the model in a single sort of interaction or inference. And, when we started this whole exercise with just GPT, you know, three point five eighteen months ago, the so so the thing that kinda started the ChatGPT wave, the token context window was more or less, I think, 4,000 tokens.

So you kinda give it like a blog post to analyze. In the past, you know, week at Google IO, Sundar announced a 2,000,000 token window model or version of Gemini. So think about that. That’s about a 500 x improvement in eighteen months of how much data you can give the model or get back from the model. And and there’s just nothing in technology history that grows at 500 x every eighteen months. It’s just never been been happened you know, it’s never been been seen before. And that is the kind of pace of innovation that we’re able to see in AI right now.

To be fair to maybe like a Moore’s Law, like religious, you know, zealot, this is a sort of a different a different variable than than what Moore’s Law is, which is about the density of of your chip or the performance of your chip. So so this is more about model algorithms and how we’re actually utilizing these models. There’s actually a different law which is sort of the GPU performance, and I think there’s a sort of a, you know, Jensen almost kinda created a law around this.

But we have seen faster than Moore’s law improvement on GPU performance. So the good news is we get a little bit of a reset on you there’s know, been a lot of talk on, hey. Is Moore’s Law slowing down? And GPUs kind of are are giving us that, you know, maybe next tailwind of compute performance improvement, which is this, you know, miracle that that that we now get to deal with. But I would say I I’m seeing no signs of AI models slow down anything plateau in terms of the innovation curves that we’re seeing.

Yes.

You know, what was interesting is last year when six months or so after the emergence of of ChatGPT, there was this mad rush for basically everybody sort of believing that the paradigm of software was you would just kinda chat with all of your applications. And so everybody launched these these chat interfaces for for software. And that was that was good because you you sort of had to have, you know, some way to get started in this movement and to see what the use cases were. If we’re really honest about about sort of the chat, you know, use case, chat is more of like a UX paradigm shift.

It takes, you know, what used to be a graphic, you know, graphic user interface and turns it more into a a command line, I’m gonna talk to it. But, like, at the end of the day, you’re kind of still accomplishing the same thing with with software in a lot of cases. You can get to information faster, but if all we’ve done is move the user interface to a chat interface, that’s sort of not taking advantage of it, obviously, the full potential of AI. So I think last year we we sort of saw the chat interface, you know, wave where, where where that was the initial, you know, kind of instinct of of software providers.

And then somewhere around, you know, middle of last year, tail end of last year, this idea or at least more established idea of AI agents, you know, has begun to emerge. Which is instead of sort of talking to AI to get information back, what if we could talk to AI to have it do something for us and really kind of complete the the underlying task that that we’re we’re actually trying to accomplish as opposed to just get information to do the task?

Everything I just said is exactly how you would have pitched RPA for the past decade. Everything I just said doesn’t sort of negate the need for for what RPA would be in the enterprise. The challenge with RPA is, you know, RPA is relatively frail. It’s often sort of that’s like looking at your computer screen and and and performing some kind of rote, you know, routine actions. It doesn’t handle variability very well. It doesn’t have the the level of intelligence that you now have in AI models.

So I think RPA actually gives you a little bit of a of a early preview of what becomes possible when you could apply more general intelligence, maybe not not not sort of full AGI, like a general intelligent model to a large number of of business tasks. So the big breakthrough is what if we could go from a world where software is something that you or I use to get our job jobs done faster or or it enables us to do our jobs to where software is something that you or I use to basically farm out work to AI to go do.

It’s a it’s a sort of real shift of how we think about software and and the role of information and intelligence in our organization. So the best, you know, examples that are emerging now are I could have an AI that is my outbound sales rep, or I could have an AI that that basically does full testing of my of my product as a quality assurance, you know, engineer, or I could have an AI that responds to all my customer support tickets. So when you think about now an organization that has, you know, effectively AI agents that augment the labor or the the people in the organization, they’re not copilots.

They’re literally autopilots that are are going out and doing work for you. That’s a pretty significant revolution in not only the software space, but literally how you’ll run a business in the future.

I I think org structures somehow seem to stand the test of time on all forms of disruption. So there’s some good books in like, go back, I don’t know, eighty years at this point, seventy years at this point that basically, like, codified today’s org structure, you know, kind of, format. And it has survived the the test of the Internet, of remote work. So so I think org structure stays the same. I think what happens is is you now have the ability to sort of drop AI labor into any part of that org chart.

And in fact, in many cases, augment whatever you would normally have had people go do. Now you have AI go and and swap in for people where you need more support.

Yeah. Yeah. Sorry. Sorry. Sorry. I mean, to be very I mean, yes. At that level of specificity, you might have certain roles that have now turned into AI roles. But, you know, the classic example would be, let’s say, customer support. My general view is that you’d have an a a layer of AI labor that handles the first first line of defense on your your inbound customer support. That will still eventually escalate to the human side, which, okay, I’ve gotta take on the more complicated processes. Maybe I’m doing things that are a bit more proactive with customers that have, you know, that that have signs of issues.

Workday might look somewhat similar. There’s just now an extra layer of AI that has been added to parts of the org chart. And then similarly, you know, if I have an outbound sales rep, they will generate leads for then the inbound sales rep that has to go and basically pursue these. So so what we’ve done is is really kinda just shift the type of work that that team was probably doing previously in in most cases.

Yeah. So so if this is a question on then total number of people in the organization, I think given a couple interesting questions. So so if you take a mature company, let’s just say let’s say Box basically, anywhere within Box, if we can make that particular function more productive so if I can take a sales rep and have them sell 20% more, or if I could take a customer support rep and have them handle five times more tickets because because AI is helping, At least so far empirically, every time that we’ve done that, we will actually reinvest in that area of the business because because there’s simply just more work to be done than we’ve ever had people to be able to go and do.

If I can ensure that an engineer at Box could develop 30 or 50% more code, we would actually reinvest that productivity gain back into engineering to build even more software. Some companies, I think, will choose to just reinvest those dollars and that efficiency. And even in the customer support case, if we can get rid of some percentage of those frontline customer support tickets, we’re gonna take that that that the time of those people and have them turn into customer success where we’re gonna actually go and, you know, reach out to our customers more proactively when when they have issues or when we think there’s an opportunity to use the product more.

So that’s that’s sort of one side of the equation. Then I think you have an interesting example, which is take a company that starts from scratch. They’ve got one employee. It’s the founder. What does that company sort of look like in the future? That actually might look like a different company. Now I I would actually argue that they they will actually if if AI is sort of doing what it’s supposed to do, I believe that company will actually hire more people over time on a like for like basis because what they would be doing is that one or two person company would hire the AI outbound sales rep, you know, software company.

They’ll actually get more leads faster than they would have as an alternative, you know, kind of way they would run the business. They’re gonna have to now hire sales reps to handle those leads, and they will literally be scaling more effectively and and probably faster as an organization. So there’s a lot of belief that, like, well, you might have these solo companies that, you know, become multibillion dollar companies with just one person or, you know, three people and and a bunch of AI. I think that will be done by virtue of, like, everything happens in the world multiple times.

But, like, I think that’ll be a relatively novel way to run their company. I think most companies would take the AI efficiency gains, and then you’ll end up hiring the functions that eventually need to be serving all the new growth that you’ve just developed because of AI.

I think there’s two types of spend happening. There’s probably the bulk of the dollars that you see in the headlines, I would argue, are likely experiments where, you know, you’ll have some kind of hit rate, success rate, that that happens. And then those experiments graduate into production spend. We just don’t have a an accurate kind of pie graph yet of what’s in the production category versus what’s in the experiment category. It would be probably too generic to say it’s all experimental, and it’s certainly not accurate that it’s all production.

The exact sort of split of those two things is is sort of hard to diagnose at this point. I’ve just been on the the road. We’ve done maybe about a dozen or so AI events throughout The US, you know, in the past quarter. The vast majority of companies have a meaningful number of AI experiments happening with with lots of different, you know, kind of areas of their business, lots of different applications. But the vast majority also have areas that are in production already. So unfortunately, it all would get lumped into the same category of AI spend by the time the CFO, you know, gets the AI bill.

But we we are seeing real production and and lots of experimentation.

We’re literally seeing both just simply because of the the way that these products get priced are so are so varied. Sometimes it’s included in your SaaS spend and sometimes it’s like, no, here’s your consumption of your AI usage.

The good news is that everybody’s experimenting with every version right now. So we will eventually see where we where it lands in six or twelve months. I’ve certainly had conversations both with customers and then startups also selling customers on every variety. So do you price to what value are you delivering? How much would that workflow have cost you prior to AI? Oh, it’s a $100,000? Well, well, we’ll charge you $50,000 for it. That one’s tough to scale because everything’s bespoke. So let’s kind of throw away the value, the the kind of bespoke value oriented approach.

Except for maybe like a Palantir or something, it’s very, very hard to kind of make that model work. So eventually, you’re gonna have some unit that we all agree with that you can go off of. So is it sort of a consumption, you know, model of what is your unit of volume? Is it customer support tickets? Is it leads that I generate? Is it emails that I write? You know, we’ll have to figure that out.

I would imagine I can sort of sufficiently say, hey, AI, generate leads for my business or answer my support tickets or review my contracts or process my invoices. If I can do that, then, you know, realistically, you will have you’ll have kind of category, you know, winners in a large portion of sort of job functions that today exist, and there’ll be AI versions of those functions. This is, again, one of these windows where ten, twenty, thirty, fifty companies will get started that were like the window in ’2 in the mid two thousands where, like, all of today’s SaaS companies basically emerged in, a five year period, essentially.

And we’ll have that for for basically AI jobs where you’ll have the AI security engineer, the AI customer support agent, the AI marketer. You know, lots of companies won’t work just for the kind of typical reasons, but we will have a landscape of basically labor that you can get from AI. Then there’ll be, like, really interesting kind of second derivative effects, which is, okay, you know, how do you manage AI labor? Like, right now, you know, when you want to manage lots of software, you know, you implement Okta or you implement a security tool.

Well, it’s kind of a crazy world where all of a sudden I have digital labor, you know, from a variety of providers. Do I need some way to kind of organize them, make sure that they can interact well, have guardrails around what they can do? You could almost imagine like a workday for AI. Like how do I actually organize what all this work is? How do I actually make sure that it’s organized well, you know, versus the what the humans are doing also? Lots of incredibly interesting kind of downstream questions in this world that that we’re only only scratching the surface of.

The cool thing is this. This will be an epic battle of existing incumbents with existing data workflow and and customers that will add AI into their platform to deliver a set of services. That will work some percentage of the time. And it’s almost, like, hard to know in advance which categories that works better or worse in. By default, that will work a meaningful portion of time. Like, when Facebook needed to get into mobile other than Instagram, they basically could go could get into every mobile category that they needed to when they chose to do that.

Only the categories that they didn’t pay attention to did you sort of see these emerging, you know, new players in the in the mobile space. Similarly in AI, I think I think there’ll be some categories where incumbents have the natural advantage, but then there will be either blind spots that those incumbents don’t kinda go after, or there will be things so different from their existing business model in kind of classic innovator’s dilemma fashion that they just don’t understand that they have to go after that that particular problem.

It looks like it’s not like a an actual threat or it doesn’t seem to be solving the same problem. And then all of a sudden overnight, you’re like, oh, the business is now over because we were disrupted by the AI version of what we do. You know, if you’re a customer support company, you might sort of think like, okay, my classic customer buys, you know, seats for every every customer’s, you know, rep that’s in the business, and that’s the business model they have. This AI agent thing looks too different from that business model all of a sudden.

And then even what you build to manage the AI agents to do customer support will be different from what the the kind of, you know, typical customer support software company would be building, then all of a sudden you get disrupted three years later and nobody needs customer support software. That that is sort of not gonna happen exactly like that, but that would be the one disruptive angle that we could see with some of these, incumbent categories.

So there were some great memes from from that interview. Well, everybody should thank thank Sam for the clarity and transparency of their of their model. It’s actually very good when when basically, you know, the lead platform player tells you the types of things they’re gonna go after versus the ecosystem. People should be paying close attention to that and, you know, really understanding what that advice means.

I think there’s a heuristic here that that that probably works, which is no matter what, I think we agree that OpenAI is kinda telling us what they are going to become, which is ChatGPT is gonna be this universal assistant interaction interface. All the, you know, subsequent tools to to manage that and interact with with different AIs that kinda come together there. And then they’re gonna have an API business that will be, you know, you can almost just very quickly easily understand. It’s gonna be audio, video, text.

Like, it’s gonna do all the formats of information with, you know, kinda complete intelligence on them. So then you’re sitting around, you’re like, what startup should I go build? You probably don’t wanna do things that instantly could be subsumed by a horizontal chat interface. And you probably don’t wanna do things in the model area that might just be one training run away on their end of being subsumed by a more superior model. This is sort of like, to me, it’s the most exciting part of software.

To 80% of people, it sounds like the most boring part of software. But it’s like, you have to do the workflows that eventually a human who wants to go and run a full business process has to implement that doesn’t wanna just do back and forth chatting with a thing. To even to your tutor point, you know, language learning, I don’t think ChatGPT itself is going to become a language tutor. I think it’ll have the full capability to be a language tutor at the model level. And could you, like, you know, prompt it to being a language tutor?

It seems seems plausible, but but, like, hard to imagine it follows a multi year journey with you, you know, exactly in the way that that Duolingo would maybe do. But what should Duolingo do is probably as quickly as possible adopt whatever OpenAI is building from a model standpoint and make sure that they are incorporating that into their product. Because what I do tend to see happen way too often is that incumbents will hold on to their own technology way longer than they should because there’s some perceived, I have to own something, that in practice, the moment that that whatever you thought you had to own is inferior to what the rest of the market has access to.

You need to drop the inferior thing as quickly as possible and adopt whatever is superior, even if it means, you know, potentially supporting what what you think, you know, emotionally is sort of your long term competitor.

Probably people would like it if Duolingo announced tomorrow that g p t four o is powering their, you know, translation engine. That would actually be a positive because it would just be like, oh, this is this very powerful AI that now your user base has access to. And we understand that, like, you’re selling a different flavor value proposition than what OpenAI horizontally is going to sell. So we think that could be, you know, better for your your business. But I I have no idea. Mean, this is like it’s impossible to give dual lingo advice on this, on this podcast.

But all all I would say is I’ve seen some people bury their head in the sand being like, no, my my custom trained model is so good at this one thing and we don’t wanna give OpenAI, you know, this particular data or we’re gonna compete over time with them. It’s like you can wish all of that to be true, but if what they have is better for the product that you’re trying to deliver to customers, that is the only way that you’re going to survive this.

That’s I think what a lot of incumbents, you know, tend to tend to run into, which is the classic innovator’s dilemma challenge that that you you see.

The tweet I sent was not for anybody else other than Box employees. So for us, we’re 2,700 or so employees. We’re still enough though that we can literally get on a Zoom call. Our all hands is like everybody in the company on one call, and then we talk through the strategy of what we have to do. I think we’re in a moment where anybody reading the news in tech understands the kind of period that we’re in. I don’t necessarily know that everybody understands what’s on the line.

If your company doesn’t make it to the other end of this bridge before everything kind of breaks off, you know, you’re out of business in this new world. So that’s probably the only alarm that needs to be, you know, kind of set.

There is a premium when you’re a bigger company, which is when you make a decision, it needs to be a well thought out decision that you’re just gonna press a button and we’re gonna go execute. And you’re gonna iterate and you’re gonna learn lots of things because of the work it takes to drive that alignment. But it means when you do that, you don’t wanna follow-up a week or two weeks later and be like, ah, just kidding. Like, we gotta do this thing. When you’re a 20 person company, you just get in a room and you’re like, okay.

Here’s what I think we should do. It looks like this. It should be priced like this. Let’s go build it. Let’s test it or whatever. And then like the whole company is is sort of like fully lined up to go do that thing. You find out that it sucks and you just like quickly pivot. And like everybody was on the same page that no, no, we were just this was just clearly a hypothesis. We we don’t really know for a fact what’s gonna happen. We can just pivot through this.

In a big company, that team still exists. But the problem is you have everybody else watching that team saying, please tell us when, like, the thing is exactly the thing and we’re gonna go off and and and kind of race. You have to kind of find exactly the right moment where you’re ready to now expose the working thing to everybody else. If it’s too late, then everybody is gonna be like, well, bozos, what were you doing? And if it’s too early, then you’re gonna have thousands of people kinda go through the the sort of like, oh, that didn’t work or like, hey, that didn’t land.

Finding that balance is kind of an interesting thing for anybody on the more product development, you know, sort of side of of these kind of companies. You know, think of Gemini actually is the first round of Gemini, all the all the kind of massive mistakes that it made. It’s probably some version of this, manifestation which is like, if it was a 20 person startup, they would have put it out there. Everybody would have known, okay, this didn’t this kind of sucks at these things. Let’s go improve it.

But at a company of the scale of Google, like there’s some team working on it. They kind of throw it over the fence. And then everybody’s like, oh my god, what is this thing? It’s just because you had a separation of the people working on the whole thing from then the ultimate people that needed to like, you know, flag the alarm. And by then it was too late and and just chaos.

I do. I think last week, to me, the the thing that I think it underscores that I think honestly, Cloud Next did as well is, like, you know, Sundar has set a message to the company that AI is the number one most important priority of all things. It it’s, like, 10 times higher than every other level of everything else you’re working on. It’s gonna be, like, the new way to search. It’s gonna be the new products in cloud. It’s gonna be in Workspace. And I think that message, it was very important.

It is now very clearly the evidence of it working is now showing up. And I think Google IO is merely the another moment of it manifesting that that a company has religion now on AI.

I don’t buy or sell anything, so I’m generally long Google in an AI era.

It’s cool they haven’t kicked you out for for for your views on this.

I’m increasingly less nervous only because of what we’ve actually seen these bills sort of come up with. They don’t don’t seem as sort of progress halting as maybe what would have been rumored about a year ago. The the scariest moment to me was the pause AI kind of moment, which was, okay, we need to we need to stop the development of advanced AI for six months until we kinda, you know, figure something out. It was like, guys, like, it’s been, let’s say, a decade of of us all as a community talking about AI.

If you think that an an extra six months is all it takes for us to have some kind of alignment on what is the doomsday AI AI gonna look like, what is what is sort of dangerous AI, what is less dangerous AI, six months is not gonna solve this. There are very fundamentally different philosophies in the land of AI that are irreconcilable. They will never fit together. It’s just it’s okay. It’s great to have actually a dynamic set of perspectives. They will not be able to ever be fully unified.

And so I thought that that was going to really kind of gum up the advancements if, you know, governments started looking at pause AI and they’re like, oh, even the tech community wants us to to stop this thing. And that that was what I was kind of most nervous about in that period. Since then, if I look at what what has legs, what where progress has been made, it’s often more surgical AI regs. So, okay, we have to deal with with copyrights and and data training.

We have to deal with IP protection. I think those are important conversations, actually. I think that that our IP law did not anticipate a world of AI, in the way that we have today. And I think it’s I think it’s a good conversation that that is is important to solve. There’s another conversation of what is the national security issues with super intelligent AI at some distant at some point in the future? How do we wanna make sure to protect against that and regulate that? And I sort of like that this is an open conversation that we should be advancing.

Yeah. I mean, actually, I think I think being an incumbent RPA vendor is is actually a great spot. Because if you already are talking to customers about literally automating business processes and workflows, and there’s just a better way to do that, and there’s nothing in conflict with their business model. In fact, if anything, actually, they they probably were the first to have more of a consumption oriented model for automation. So I I think I think you could be very bullish on RPA vendors right now. I do think it means that more players kinda get in and around the space.

The thing that will 100% guaranteed happen, like, in five years from now, this will be the most obvious thing of all time. But when you look at, like, RPA, you had to be, you know, a relatively deep expert in in RPA. You had to be like a mid size or large enterprise or or kind of developer oriented, you know, kind of individual. The total size of the market was basically arbitrarily or artificially held back by just the complexity of the legacy approach. So if AI makes it 10 times cheaper, faster, and easier to automate workflows, then it stands to reason that the market will be substantially larger.

It could be a 100 a 100 times larger at the end of this whole journey. So what will happen is the market will get much larger. Traditional RPA vendors, they execute, should take a good portion of that market. And there’s gonna be a new set of entrants that will bring automation and intelligence to their platforms that take out parts of of the new market that emerges. That’s obviously what we’re focused on, which is, hey, what about all the parts of the business that you never automated before?

You know, if you go to most companies and you say, show me where your digital assets are, all your marketing materials, they’re gonna be like, okay, it’s in all these folders, you know, inside of of my hard drive. And, what if you could just like go in and and ask a question of like, show me all of the, you know, you know, red dresses and it just like instantly came up? Well, that is what you’ll now be able to do with AI. What if you go to your contract repository and instead of it being just a bunch of folders with contracts in them, just say, hey, what contracts are up for renewal next month with this with this, you know, particular clause in it?

And instantly, now know, you know, what areas of risk does your business have. So these are things that you just could not have done before that AI now lets us do on top of all of our data.

I mean, probably the long pole of all of AI right now is not gonna be the technical breakthroughs. It’s going to be the implementation and change management on the human side.

Do you literally mean the Accenture type services?

Services are almost the leading indicator to compute. Because you need services to implement the thing that then eventually is sort of on autopilot. So I think, I would take the bet on AI services for the next five years unquestionably. The amount of dollars that will go into the change management of systems, the implementation of the technology is is gonna continue to be massive. On the other hand, the other thing that goes along with that though is a bunch of the AI stack. So the actual GPUs, the data center build out, that as well.

At some point though, the curve, if AI is is as meaningful as as I believe it is, and I think, you know, so much of of tech believes it is, at at some point, the actual AI, the software services of AI, the the infrastructure services of AI will eventually exceed the human services on the implementation simply because now once it’s in production, you don’t need that same change management ten years later. Like, it’s just literally running. Like, the amount of money we spend today on our cloud infrastructure vastly exceeds the amount of money we spend maintaining our cloud infrastructure versus five years ago when we were first moving more into the cloud, our services were higher than our infrastructure spend.

To me, there’s just interesting downstream kind of consequences if everything plays out as as it it should on, like, on paper. So I’m fascinated by the idea of, you know, what SaaS did was it made it so I have a friend who has, he sells balloons online. It’s actually like not a bad business. It’s like a like it makes real money. And he sells balloons online, and I don’t think he would have started the business if Shopify didn’t exist. The existence of Shopify made it so he could, like, be like, oh, well, I had this, like, random idea.

Let’s just see if it It lowered the barrier to then going out and and basically starting a business. And I I know that Toby has thousands or tens of thousands of these types of stories. So if Shopify caused businesses to get started because it lowered the barrier to being able to sell online, if AWS caused applications to get started because I was like, oh, I could just build an app and and run it in the cloud. Don’t have to think about servers anymore.

If Stripe caused businesses to get started because I don’t think about payments anymore, then in a land of AI agents, you can almost similarly be like, well, you know, somebody literally one day could just be like, well, maybe I should, like, start this idea because I’ve lowered the barrier to, again, getting customers or supporting customers or, you know, testing the software. Like like, I now have AI labor that can let me now scale my my startup idea or whatever. And I think there’s an interesting global democratization that could happen where you could be a company based in a place that, like, did not have this type of talent for a a business being started.

I’m I don’t know which country I should pick on and and as a made up, you know, kind of country as this problem. But, like, if you’re two or three people and and name your random country, does that country have like the SDR sort of infrastructure that that got built out in parts of Europe or or or The US because of the SaaS boom? Probably not. But like now, I go and I can just like literally like hire an outbound sales rep, you know, with AI.

So think about what is that gonna mean in terms of where companies get created, how much you can scale up. I think we’ll just see a a boom in, all new ideas begin to emerge.

Yeah. I I think you will always see this a standard Power Log dynamic. I think if we were having this conversation imagine having this conversation twenty years ago exactly and being like, wait a second. If the iPhone lets you have any software and Amazon lets you build any software, aren’t we just gonna see like tens of millions of apps? How will anybody even, you know, know what to find or discover? And I don’t know. The world the world magically just kind of figures out this stuff.

Like, the the things that don’t work die off. The people don’t pay for the outbound emails anymore. The things that do work take off. Creative destruction is is an incredible thing that’s alive and well in the world.

I I think we have ways of defending against this, but, like, you can it doesn’t take, like, that much imagination to be like, oh, you know, one of those, like, robot dogs with, like, a gun and a multimodal AI. Like, wow. Wouldn’t want that thing running around. So so I I think there are real reasons we should sort of pay attention to some some of the more dangerous use cases of AI. I just think we have, in many cases, sufficient legal frameworks for addressing those things.

And then for any net new one that we come up with, let’s actually have regulation to go and support that. But what I’m I’m less worried about at the moment is probably the more fringe extreme things of the AI sort of self replicates, jumps over outside the data center to another data center, and then self propagates. And I’m less nervous that that is is sort of on the on the docket of events.

We’ll see.

Definitely delegation. I like to get my hands dirty with all aspects of the business. And so it was it was very hard to kinda let go of parts of it to to other leaders.

I I cannot wait until we’re a $2,000,000,000,000 company, and I will be able to I will be able to go back into that mode. I’m in a brief period where I have to delegate to keep people happy, but eventually we will we’ll be back to that land. I think the the real lesson is now is you choose the areas that you you need to exert that level of involvement. So in places like, let’s say critical areas like AI or end user experience and and some product, you know, strategy, I still kind of revert back to, you know, early startup self.

But there are many areas where just honestly either because of the amount of hours that the job takes or just the fact that like you want to be able to bring on great people that that are motivated to go execute, delegation is actually extremely important.

In life or in AI?

My read on this is you have, you know, one of the best entrepreneurs of our time that has basically all of the right underlying resources for AI. So lots of compute, incredible engineers, billions of users, I. E. Lots of data. And I think he is extremely motivated to have a platform that the Internet builds on and participates in. And I think AI is sort of his moment to really kind of reestablish the dominance that I think we remember Facebook having in the maybe late two thousands, which is like we were all gonna build on the social fabric of the web.

Some of that happened, some some didn’t exactly happen as planned. And I think AI is now this moment when actually he can provide a real alternative to more of the, you know, relatively closed approaches to AI. And he’s sort of set up extremely well to to be that kind of counterbalance in this, in this space right now. So so I just think it’s like all of the stars aligning for his particular both philosophy and and resources.

We had Jensen speak at Box six years ago, and Gen AI was not a thing, but AI was a thing. So so I’d say Jensen having access to, you know, basically the start of the supply chain of AI is such an important perspective on where things are going. I mean, you if you could just look at his invoices to and from TSMC, if we could just see that data, you would kinda know what to go build next. His window into, you know, five or ten years out in in tech is is unparalleled right now.

Earlier in our journey, I would have focused more on cash flow. I I have become a little bit of a of a religious around cash flow. I think owning your own destiny as a company is important. I think caring about every and inspecting every single dollar of spend in the business is very important. In sort of very loose capital environments, it sort of gets forgotten about or people don’t really care about it. But actually, I think it helps you build a better business because you apply constraints that force better decisions, better strategy, better execution.

So I would have done that years earlier than when we ultimately focused on cash flow.

It helps that my wife is equally busy. So I don’t get yelled at as much as probably as possible for for for my own business. I mean, you you just Probably all the standard things. You just You carve out the date nights. You have weekends for family time. And then you just ask for forgiveness a lot.

I’m I’m bullish on Apple for a variety of reasons. You know, institutionally how how the the company operates. But there’s some meme probably of, like, why is Apple not fully in AI yet? First of all, if you look at the most immediate threat that AI poses for for companies, I don’t think Apple is is immediately threatened by AI. Most of my AI usage is on an Apple product that I have often paid money to Apple to be able to use. So, like, if I so, like, just like on the most immediate basis, all of my usage of AI is not threatening to anything that Apple does and is purely contributing even more to their platform dominance.

But now let’s go on the offensive side. So that that’s sort of on the defensive side. Like like, you have to first assess, you know, does this new technology hurt a business that I’m in? And I I think the answer is largely no for Apple. They’re not selling something that AI is is sort of directly threatening. Then you you flip to the offensive side, which is does AI help the business that that Apple is in? And I think it’s unquestionable that if you could turn this thing into basically a command center, I just say something to AI, and it does a thing in the world, gets me an answer to a search query.

You know, hey. You know, what what’s a a good, you know, way to make dinner tonight? Whatever whatever you you you do as a on personal side, you know, instantly they’ll just do that. And then if it really turns into a thing of like, hey, book me this flight or call this Uber for me or whatever, that’s just an infinite amount of of new types of transactions that they can facilitate that they are not currently monetizing in any kind of meaningful way. Now who knows how they actually monetize those things, but I think it turns your phone basically into into your your task, you know, automation engine, this intelligent, you know, sort of device.

And so if you have a billion plus people that are using this device for now a new set of things that they never used it before, it’s all through your interfaces, your APIs, or your software, I think that puts them in a very powerful position. So then the only question is like, oh, well, so why is it taking kind of quote unquote so long? I would actually argue that any of the any of the kinda quote unquote either delays or slowdown doesn’t amount to anything. The AI that I was using a year ago is totally different than the AI I’m using today.

This this space is changing so rapidly that it is way better for Apple to enter the moment when they, you know, believe that they’ve built the right user experience, when the technology is sort of stable, and when it’s high enough quality, we will all use with the thing that they launch whenever that point is.

But that’s that’s my thing. It’s like, there’s not that’s not even an or. Why is that an or? Like, what does it matter to you? What is the underlying engine in Siri? I don’t care. I want to be able to click my phone, ask a question, get something done. So so I don’t even see that as an or. I see that as just a a natural sort of technology partner supply chain thing. And who even knows what they’re even doing because it’s all rumors? I don’t see that as as any weakness on Apple’s front.

I see that as a as a consumer. I just want the best AI to be on my device.

There’s there’s really metrics like, you know, could we we we just passed a billion in revenue. We’d like to get to 2,000,000,000 as quickly as possible. That’s an important milestone for us at at, you know Can I can

is sort of why I am the lesson for everybody’s future is eventually, at some future point, you will be measured on on a set of metrics that are much more boring than ARR multiples. And they will be free cash flow multiples or or whatnot. So, you know, we are we are at the later stage of that journey. Now, four x is clearly way way too undervalued. It you should be in the kind of, you know, six or eight x range. We had a period probably the last couple of years where the kind of core business slowed down and we’ve been launching a, you know, sort of second act, as it were, or third act within, you know, both AI, workflow automation, more business process.

That takes time to get to scale before obviously the the top line growth, you know, can slow down. So so we’re kinda in the midst of that transition right now, but these are the things you deal with as a, as a public company.

No. No. It’s fine. So our next set of goals, get to 2,000,000,000 as quickly as possible. Market cap, of, you know, whatever is associated with that. You know, I don’t have a specific one specific thing as much as kind of back to what we were talking about earlier on AI. You know, if you think about the information that goes into into your content, your contracts, your marketing assets, your your financial records, your invoices, this is basically, you know, the closest thing that a company has to its digital memory, let’s just say.

And AI now has the ability to actually tap into your digital memory as a company. The thing that that over the next couple of years, and so let’s say five as an arbitrary point, is really enabling just the world to reimagine what they can do with their data. And and what if every company had infinite digital memory to make better decisions, to automate more processes? You know, think about the new employee that has to be onboarded that doesn’t know anything about what the hell, you know, the organization’s doing.

Instantly, that’s now what they can tap into. So so we think it’s just a profound moment for for how you work with your information in the cloud.

That’s right. I I who wants process? So we’re good.