How SaaS can survive the AI revolution, with Aaron Levie

Stripe is built for whatever comes next because AI isn't just changing business, it's evolving fraud. Stripe Radar detects fraud with the fast scale of Stripe's 1.9 trillion dollar annual transaction volume. In fact, when a card is used, there's 92% chance the Stripe network has seen it before. That means emerging fraud patterns are detected sooner, blocking bad actors, welcoming good customers. Learn more about fighting fraud with Stripe at stripe dot com slash radar.

The business world is moving faster than ever, and when change hits, we need to learn in real time. On Rapid Response, you'll hear candid conversations with CEOs and leaders making tough calls about AI, human talent, responsibility, and the bottom line. How they navigate uncertainty, pressure, and high-stakes moments. I'm Bob Safian, former editor-in-chief of Fast Company, and I'll be your host as each episode breaks down what you need to know right now. You can find Rapid Response wherever you get your podcasts.

On Masters of Scale, iconic leaders reveal how they've beaten the odds.

Asking really strong questions is a superpower.

You want to show up with something radically different and how they've grown companies to incredible heights.

The greatest rewards always come from the greatest risks.

Let's hit the gas.

Airbnb, Zillow, Microsoft, Liquid Death, and more. Hear from the founders who've changed the game. It's anything but business as usual. Find Masters of Scale on Apple Podcasts, Spotify, YouTube, or wherever else you get podcasts.

Well, for us, everything's at stake. I would say that other than that sort of very early founding period, maybe the first year or so, it's never felt as much like a startup as it does today. We feel like we're very much back in that early formalization period, mostly because of the size of the opportunity.

Aaron Levie is co-founder and CEO of Box, a SaaS company now worth over $4 billion. They've been around for two decades, so why is Box suddenly infused with the frenetic energy of a startup? Two letters AI.

Every single day there's a new model release, there's some new breakthrough in how you design and build agents. So it it really, really is is critical that we are moving at this, like, very, very fast-paced way. Everybody working on AI feels that type of pace and intensity right now.

Even as a successful founder and CEO of a major public company, Aaron finds himself back in the trenches. His company is a leader in developing cloud-based enterprise tools, so he knows what it means to strike while the iron is hot. So on this week's episode, we ask, is SaaS dead? As one of the biggest figures in SaaS, Aaron makes the case for how companies like his can ride out the AI transformation. I'm Rana Al-Kalyoubi, and this is is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution.

Aaron, welcome to Pioneers of AI. I'm so excited for our conversation.

Thanks for having me on.

I wanna kinda go back to your origin story. So it sounds like or it feels like you've wanted to do startups forever. When you were an undergraduate student at USC, I guess you explored with, like, starting a search engine and a real estate portal and a whole bunch of ideas. And then it was an internship at Paramount that led you to starting Box. So tell us that story. And I guess, were you fixated on starting a tech company, or were you looking for a problem that you're excited about and wanted to solve, or both?

Yeah, I I got enamored with technology fairly early on. So in the early years of high school, I had a couple friends that were building software and, uh, building kind of internet websites. That piqued my interest quite a bit, and so I s I started building lots of w lots of websites and and through that created a bunch of projects with with again, kind of various friends, sometimes on my own, sometimes with what eventually would become one of Box's co-founders.

I was just trying anything that w- that's that seemed, you know, interesting and a good use of the internet. And, um, none of those ideas really kind of panned out, but went to college and tried tried to explore another kind of passion area, which at the time was film. Still still remains a a a kind of an area of interest, but thought maybe I could combine digital technology and media and and do something with that. So, got some internships in the entertainment industry, and then I was an intern at Paramount in 2004.

And at the exact same time, I had to do a project in college to kind of identify a market opportunity. I chose this idea online storage, sort of through just the a random set of events of working from a corporation and sort of seeing the, you know, how sluggish legacy software was. And then in school, just like emailing yourself files to move between computers, so kind of a a few stars aligned and and eventually decided to study this space of okay.

Well, there had been companies that would let you store your data on the internet, but they all kind of died off in the late nineties. They didn't really innovate, they were sort of these zombie products and companies, and it felt like right this m this moment where where you could finally sort of deliver on this promise of. You could put your data in the in the cloud. It wasn't called the cloud at that point, but you could put your data online, store it securely, access it from anywhere.

Browsers were getting faster, the internet was getting faster, storage was getting cheaper, and so that was kind of the the general thesis that we had. And, uh, started building that idea out, and then we launched it in early 2005 and then just sort of started growing from there.

It's amazing. So we recently had Mark Cuban on the show, and yeah, you have your own Mark Cuban story. Uh, I guess he gave you 350 k check, but I'm most curious about, I guess, he advised you to never hedge your bets in a startup. So I would love to hear more about, like, h- how did you implement that advice at Box and do you still live by it? How did you know him, by the way?

kinda probably how everybody gets to know Mark Cuban. So you he it usually starts with an email, uh, from you to Mark Cuban, and then from there, a relationship sort of blossoms. So it was our between our sophomore and junior year of college. Box had been around for about four months or so. We were trying to just tell everybody about it and also raise some money, uh, uh, as well. So Mark was was sort of right at this intersection where he had a popular blog, so we we wanted him to just maybe he could just tell people about it.

And we also were raising money constantly. We were pitching everybody we could find, everybody was rejecting us, so we had, like, a very, very low hit rate. And but Mark, he is passionate about entrepreneurship, he's fine to kind of throw a hail mary for a crazy idea and bet on crazy founders, and we kind of qualified on all of those things. Like, the you know, this idea had been tried but failed miserably through the nineties, so it was like, on paper, it would seem like a bad idea.

We were brand new entrepreneurs, at least in the in the kind of, you know, ecosystem that that, you know, real companies were coming from. So it was very much a kind of a hail mary for us to to pitch him and and for him to even invest. And but he made this investment.

love that story, by the way, for all the founders who are listening, right?

Yeah, I mean, the the takeaway there is just pitch everybody and just don't give up. I look back and I can, I can sort of figure out why it was Mark Cuban that invested, but but we pitched another thirty people or fifty people, and everybody else said no, except for a couple other people, and it could have been somebody else and not Mark Cuban. So you just you just have to pitch everybody and just like, and then, you know, if you are so lucky that you have multiple choices to pick from, then pick the people most relevant to you.

And Mark actually would have been very high on that list just due to his background, but we were unbelievably thankful that that he made the investment. That led to actually us dropping out of college and betting on the company full time and pursuing it full time. But within about sort of six months or so of the investment, a around that time period, we were trying to decide a couple different business model options and even technical options.

We were split between a couple different approaches that we wanted to take. And, uh, he basically said, you know, as a startup, just don't hedge your bets. Like, y- you're it's either gonna work or it's it's not going to work, but it won't work if you try and do both things. And and so from a even like a game theory standpoint, that's sort of the best advice. If you have a startup with with very limited focus and attention and bandwidth, if you try and do two things that are of opposing sort of directions, you're guaranteed to lose.

If you do one of those things, maybe you only have a fifty fifty chance, depending on which one you chose, but that's your only path to having any ability to to get a return. Because you're just gonna be too diluted, your energy's gonna be split in too many directions. And so that that was the advice he gave, and we live by it today. Every time that I see a situation where we it feels like we're hedging, which is like we can't quite make a decision, so we wanna try both things,

Channel Mark Cuban.

we just fully live by it. Now, that's that can be different from keeping your options open, and so it's important to not kind of confuse the two. So we often are keeping our options open, where we will have an, uh, uh, uh, maybe a technology framework approach that's sort of flexible, it can move in different directions depending on the trends that we see in a market. But if there's something where you really need to make a fundamental decision, we do not parallel track it.

We we make a bet, and we generally go big on that bet. And so far, our hit rate, I think, is I I think pretty good, and it's avoided a lot of sort of disasters that guaranteed would have happened if we had tried to go both directions.

So what's an example where parallel tracking kind of multiple options would have been disastrous,

Yeah, so once we pivoted into the enterprise market, which we did relatively early on, a couple years into into the business, and it's a a great example of just not hedging. We were all in on enterprise. We we had a lot of customers that said, Okay, could you deploy an on-premises version of this software where we could just run it in our own data centers? And this was a time where the cloud wasn't 100% obvious.

Um, there were a lot of enterprises, especially in banks, pharma companies, governments, that didn't want to have a cloud version of technology. They needed to be able to manage it and control it themselves. So a lot of companies at the time were actually giving enterprises on-premises systems at the same time as building out cloud products because they they wanted to be able to win the deals and not have to convince the customer to go to the cloud.

We made the the bet that architecturally, it was gonna be way more important to be fully in the cloud because as new each new innovation happened, mobile devices or eventually AI, you need the ability to have all of your customers have access to that technology on day one, the moment it exists. When you have customers on on-prem environments that are managing their own systems, you're at the mercy of their upgrade cycle and how often they can keep their technology up to date.

And then what's funny is actually the end users end up blaming the vendor. They don't blame their internal IT systems. They will they will think to themselves, Oh, that Box product isn't very easy to use because it doesn't work on my phone. And so what we realized was we don't want to be in a situation where we have all of these customers that have fragmented deployments of our software. We need to be able to be in control of all of the new innovation that they get.

And so we basically said, we're not gonna hedge our bets, we're betting fully on the cloud. It will mean and it did mean that we will lose deals along the way. There'll be lots of banks that we don't win, there'll be lots of pharma companies, there'll be lots of government agencies that we don't win. But guess what? Years later, certainly as all of our cloud customers kind of grew, they got all of the benefits of using the most modern features and technologies that were available.

But probably the vast majority of the companies that said no, we have to stay on-prem eventually set realized they have to go to the cloud. And we had now a better architecture to have them move into. And all of the the players that were doing the on-prem software were the software that those companies were moving from. So, yes, you had to be much more patient, but if you were right architecturally, you end up in a way better spot, assuming that you kind of stick to your principles.

So the general advice there is for, you know, any startup, certainly don't hedge your bets, but absolutely don't change your your technical principles, because if you're right, customers eventually will move to your side. If you're wrong, the business wouldn't have worked out anyway, so it doesn't matter.

Yeah, that's great. All right, I read somewhere that you were a magician in middle school. I don't know, are you still a magician? Do you still do magic?

I can s uh, occasionally, um, uh, produce a a magic trick.

Well, actually, at MIT, one of my lab mates, Seth, Rafael, was a magician, and he studied the s the emotions and science of wonder, which I thought was really cool. But anyway, the whole, I guess, premise of a magic trick is that you are distracting the audience to look at the wrong thing while you're here doing the real trick.

Well, first of all, you're not supposed to you're not supposed to know that. Okay, okay, all right, That's like our little secret. So, yeah, you don't have to tell people this.

okay. But anyway, what I wanna apply this to AI, so

Oh, yeah, great.

yeah, what? What is everybody distracted with? That is like the distraction. It's not, like, where the real value of AI is being created, and where do you think the real value of AI is being created?

Oh, interesting segue, cause actually we do use magic. I I use some of these principles in the actual product.

Cool, okay, say more.

Well, there's a lot of stuff you try and do where you you need to buy yourself some time in the in the product experience while the the system is working. It's not a classic misdirection, but it's ways of sort of the system is sort of thinking, and how do you make that not not very boring, but you actually give people a a real experience through that process? And so, uh, I thought maybe there's a chance that that's where you're going, but yeah, I on the overall space, I think that I I'm such a deep kind of enterprise person that my general belief is enterprise AI is is where most of the power of AI will will be delivered.

AI at the end of the day, for the most part, minus a couple categories, I can I can argue a few for a few categories that are kind of consumer oriented. I think, really, it's a it's an enterprise industrial technology. It's much more akin to kind of our classic industrial automation systems, but for knowledge work. So if you saw a a gigantic, you know, forklift for the first time ever, you wouldn't immediately be thinking that's a massive consumer market.

You'd be like, Yep, this is gonna be in warehouses. And so I think of AI like that. I think there's a lot of fantastic consumer tools, search and and finding information and maybe therapy and a few of those things. But I think the big market is the big TAM is on the enterprise side. And then I think I'm probably just a pragmatist in the sense where I think you're gonna have model progress that just continues to to go at an incredible rate, which will just keep happening.

But model progress will always have a capability overhang effectively until you have the the workflow systems that take the models and put them into the business process. And so I think the the part that maybe some people have wrong is that that will take longer than than maybe some of the most optimistic or bullish AI people believe. I think the change management and enterprise workflows just takes longer than than most realize. And I think that will cause a little bit of cognitive dissonance because in in Silicon Valley and in tech, we will see these incredible breakthroughs and and we will be using them because we'll be coding with them and we'll be changing our workflows very quickly.

But then for the rest of the world to actually adopt those and transform a business process, transform their life sciences process, transform their manufacturing workflow, that will still take two years, three years, five years, maybe a decade. We saw this in the cloud, where everybody in tech, like within two years of like, let's say, like 2007, 2008, 2009, we were all like, Yeah, obviously, cloud computing, that's the future, we get it.

Well, guess what? It took two decades for that to actually play out across all all businesses, and still, we're still going through the migration phase. I was with a customer four weeks ago. They are the biggest believers in cloud of all time. They there's no question they love the cloud, and yet they still have massive systems on-prem that that they are only now starting to think about. How do they migrate? That is the kind of change management that actually happens in, you know, kind of large enterprise and in the corporate world.

This is gonna take a little bit longer than people think. Maybe the good news is if you're on the entrepreneurial side of this, it means you have time to build great solutions for the market. If everything could just instantly be solved by the model, there wouldn't be a lot of opportunity for entrepreneurs. You know, the labs would just have all the value. So I I think it's mostly a a a good news story. It just means that we have to be a little bit more patient with our timelines.

In a minute, we tackle the elephant in the room. Is AI the end of SaaS? Why Aaron is still betting on SaaS after a short break. The companies winning right now aren't just building great AI. They're building the infrastructure to actually run a business on top of it, which is where Stripe comes in. Stripe is built for whatever comes next, with planetary scale financial infrastructure to help you power any business model. As a matter of fact, 86% of the Forbes AI fifty run on Stripe, along with industry leaders like Salesforce and Pepsi.

Last year, Stripe users processed $1.9 1 trillion combined. See how Stripe can power what's next for your business at stripe dot com. Welcome back to Pioneers of AI. You can also watch this episode by heading over to our YouTube channel. So AI is definitely creating this kinda structural shift in how and where value is being created. And I wanna segue next into software as a service, and I'll just name it as it is. There's this meme going around that SaaS, as we know it, is is is dead.

And and I just wanna kind of unpack why I'm s like, I wanna position this and then, of course, get your point of view. So so the idea is that SaaS produces these tools that help us do work faster and more efficiently, right? Like, and and it's it's mostly kind of a systems of record, whereas AI just does the work. And the example that I like to give in, uh, is, you know, instead of creating a tool that helps lawyers organize their contracts, you are just creating an AI lawyer that is going to redline, review, approve the contract.

What what do you think about all of this? Box is is the system of record, right?

I am a believer in exactly your set of sentences. Where where people get it wrong is thinking that the agent that does the contract doesn't need a system of record to work on. And and that's the that's the part that that I think some of the narrative misses. So, like, I 100% endorse the vision you just laid out. But then the question is, well, how did the agent get the contract? How did the person interact with the agent to give them the contract or to get the output of that contract?

How did the ecosystem of other enterprises interact with the agent for the the redlining process that has to go outside of your firm? Well, all of that actually needs software. That needs workflows, that needs data governance, it needs access controls. My argument would be that that the system of record becomes meaningfully more valuable in a world of agents. Because my my general view is is that what what you basically have is like a thousand x increase in the number of users on these platforms.

So it used to be that when we sold, let's just take that that contract example, when we sold a document management system to a a law firm or a company with a legal team, you could only sell to as many lawyers or as many people were in the legal operations of that company. Now, with agents, that law firm might be doing two times the amount of work or five times the amount of work with agents. So now, all of a sudden, the system that has to maintain what agents are allowed to do, what on which data and which people can see the output of that data, and how do you manage and govern that whole workflow?

That actually increases in value in a world where there's more agents or humans on those systems. Because the the the stakes are now higher, right? An agent getting the wrong piece of data, answering the wrong question, combining the wrong information, uh, leaking data to the wrong person. The stakes of that now have gone up by an order of magnitude. And and we are only in the earliest stages of seeing what what AI security is going to look like.

What happens when you prompt inject an agent to pull out data from a system you shouldn't have access to? All of a sudden, people are gonna care a lot about their systems of record to ensure that they're both that they're up to date, they're maintained, they have the right data, their access controls are correct, their security's good, that they connect to all of the agents in the right way. So that that's sort of why I think the premise is a bit off when people talk about, well, the SaaS just sort of get consolidated into agents.

When was the moment you realized that y- y- you box have to go from this, like, being the system of record to implementing these agentic workflows? And what did you then do to bring that to fruition? And and perhaps also then give us an example of how you've worked with an organization to make this happen.

Yeah, I mean, we realized it basically probably within ten days of ChatGPT. And I I kinda look back at that period and honestly, it's a little bit like like, I probably just should post more to my own thinking process because I don't know why it had to take ChatGPT to make it so obvious what we should be doing with LLMs. But the ChatGPT moment was, like, just the right packaging, the right user experience that kinda made it super obvious what we were all sitting on.

So within about ten days, we were like, okay, if people start asking questions of, uh, of an AI system, then obviously I shouldn't just ask questions of of a trained model, I should ask questions of my enterprise information. So, like, that was, that was obvious. Now, again, to be clear, like six years prior, we actually had a project to attempt that. It just was never gonna work because until you had, like, 200, 300 billion parameter model that could be run efficiently, like, you just wouldn't be it wouldn't generate the right answer.

It'd be very domain specific. So so that was kinda like very shelved, and we were not even like, it wasn't even in our headspace. So it really required the the ChatGPT, you know, experience to manifest it again. So then we pivoted. It started small. It was like a ten, f- fifteen person team, and and we were we were focused on the experience you just said, which is like, ask documents questions. That was great. And and then the other very obvious use case that emerged was, well, if I have an AI model, it can read a document.

You can then pull out structured data from documents, so that becomes that lets you take unstructured data and turn it into structured data. That's very powerful for a wide variety of use cases. So we were kind of cranking on that. And then I think the the really big, like ten x has just been agents in the past year. And we we started really getting our arms around this idea of, well, what happens when you can just deploy an agent to go and do work for you?

And what you give them is access to any of the most important data for that work. So I want to go and review a bunch of financial documents and have an agent go and do a due diligence process on that financial data. That that's obviously a a highly valuable activity in the economy. And so can agents sort of loop through data hundreds or thousands of times for a long-running task that could take hours? That's the the kind of technology that we're now building, so that's that's the early stuff.

The stuff that's live right now that is a meaningful breakthrough for us is just give us your contracts, your invoices, your healthcare documents, your research files. And we have agents that go through all of that data and extract all of the structured data from that. So then you can automate a broader workflow, or you can get business intelligence on all of this content that you're sitting on, which is great for if you're a real estate firm and you need to know, you know, what are the clients that maybe could be saving money on their next renewal at the site that they're at?

How do you have data across all of your clients, across all of your contracts and lease agreements that give you that kind of insight? Let's say you're a talent agency and you wanna know new ways that you can either package up talent or projects or monetize in new ways. Well, what if you could read through again all of the artist information you have, all of the contracts that you have, and be able to just in natural language query all that data?

Those are the kind of things that our customers are now starting to do, and it's either automating workflows or delivering new kinds of intelligence and insights into their business process.

Yeah. So I wanna talk about business models, because traditionally, again, SaaS companies, the business model was a per-seat pricing model, right? But increasingly, with a lot of AI companies and agentic AI workflows, it's outcome-based. Like, did you complete the task? I'm also an investor, for example, in a number of companies where it's almost like they're leasing a digital coworker or an AI agent or agentic workflows. How do you think about business models in this scenario where you're not just like the organization system, but you have all these tasks that you're helping your customers complete?

I think for us, it's gonna be a a hybrid. There's gonna be seats for the user that will be a predictable price point, $10 a month, $20 a month, $30 a month, whatever the plan is that you're on. And then you're gonna pay for some volume of agents that might initially be included in that seat plan, but you do have to have an affordance for the consumption variability that that just happens naturally. So one company wants an agent to review 10 million documents, and another company occasionally wants an agent to review fifty documents a week.

They should be paying different amounts, they shouldn't pay this the same price, and and everybody agrees kind of across the customer base, you know, that that that's the case. And so I think agent pricing will be more akin to if I were to hire a services firm to to do something for me, I I know that it's gonna be volume-based in some way. And I think you're gonna see that mostly in AI agents, and I think what certainly probably whatever your investments are doing and what Cursor does and Claude Code does, Codex, et cetera, like, there's gonna be something in this space, which is you get a certain amount that's kind of included, and then and then you're gonna do

some consumption model for any capacity on top of that.

Who do you see as Box's biggest competitors? Because on the one hand, you've got all the storage companies, but on the other hand, you've got all these, like, AI coworkers, AI agentic workflows. Like, uh, you know, is it the Harvey AIs of the world, or is it who you think of as traditional storage?

We need to be the best place where you want your content to go, whether at that is at the moment of creation or at the moment of where you're kind of managing it and governing it and the access controls to it. We wanna have the data show up in any application you're working in. So whether it's a Harvey or something else, we we're we're gonna often be very complementary to those systems. But there's always this sort of moment of truth, which is, well, where, where ultimately is that piece of data going to reside?

And our brand promise, our platform promise, is we wanna be the best place, the most secure, the best governed, the easiest to use place to manage that information. So I think historically, that meant that we competed with other products that could store your data. Increasingly, though, as you have agents, then we have to make sure that we're offering the right kinds of solutions for customers to use those agents on their data in Box. But we're not territorial on what around where you use those agents.

If you never logged into Box, but you used an agent to interact with that data inside of ChatGPT or Harvey or somewhere else, we're totally comfortable. So we want there to be a thousand, you know, amazing companies that get built out, and what we want them all to do is be really, really hungry for working with unstructured data. And then we're happy to build the systems that make it really easy to get that data into those applications.

That is very cool. Can we talk about security and governance? Cause, you know, if you don't get this right, there is a risk of the AI kind of getting access to files it's not supposed to get access to. So how do you implement all of that? And I guess it depends on, you know, if you have files in Box, it depends on who the human or what function has access, and how do you mirror that?

Yeah, that's exactly right. So today's paradigm basically says, if you really think like, what do agents? what are they really doing for now? What are agents really doing? Is they're taking one person's capacity and multiplying it. Okay, so if they're taking one person's capacity and they're multiplying it, then basically what we're doing is we're giving the agents the things that that person has and then just letting them do way more on that. So if I have Claude Code, it's doing stuff on my computer or accessing my GitHub, and it it's sort of representing as me in those systems, and I've sort of provisioned access to it.

So the most important thing to get right is figuring out what should the user have access to, and making sure the agent only has access to things that the user has access to, and that the agent can only answer things back that the user also has access to. The moment you break that paradigm, all bets are off. The agent will absolutely reveal everything that it ever has access to, and you will totally be getting all of this information and IP that you're not supposed to have access to.

So the simplest example, just let's just say you had no permissions in your company, and you took all of your company's files and you put them on a server and you gave that agent the server. And an employee says, You know, what's Sally's salary? Well, the agent will just answer the question, Uh, what company are we buying next week? Well, it'll just it'll just answer the question. So that that's the extreme end, and you can kind of, you know, use that analogy for anything in the business, right?

I if an agent had access to the entire HR system of a company, it'll ask answer any question to any employee that asks the question. So if you kind of start with, you know, kind of the most extreme state, then you have to figure out, well, what things do I have to put in place to make it so the agent can only answer things around questions that the user has? Which means I have to have some degree of human-based access controls that the agent basically conforms to.

And and so things like data governance, organization of data, keeping access permissions up to date, this is going to matter, like, by two orders of magnitude more than it ever has before. And so at Box, we we build the technology that makes it, you know, hopefully as easy as possible, but makes it so an enterprise can manage all of those access controls, all that governance layer for their documents and their unstructured data.

We're going to take a short break. More with Aaron Levie in a minute. Stay with us. All right, so I want to, like, zoom out a little bit. My thesis around AI is that it's basically expanding the TAM, right? You're we're not just disrupting the software market anymore, and you've already kinda referenced this. You're kind of eating into the services industry and maybe even the labor market at large. How are you thinking about this? Would you, I mean, would you agree with this?

How? How do you think it's gonna like unfold over the next few years?

Yeah, I think AI represents a a TAM expansion depending on on your math and how you cut it, if you do bottom up or or top down math, it's at least a kind of a three to five x increase in the size of of software, maybe a ten x increase in the size of software.

Walk us through the numbers.

Yeah, I mean, it kinda depends on just the categories that that that you you end up cutting this from. But if you basically think about it as SaaS is a couple 100 billion dollar market, software, enterprise software, labor is

Trillions and 1 trillion, yeah, tens of trillions, yep.

so. If you kinda look at this as okay, could you take 10% of that, 15% of that, and pull that into software dollars, that's sort of how you end up with any of these numbers. And so you can see the the magnitude that we're talking about. Labor very large, software very small, AI lets you you kind of just pull down more of that labor. Maybe a more bottom-up way to think about it is inside of the average enterprise, usually you're spending a few percent on IT of your total revenue, like a a couple percent to maybe like two, three, 4% goes into IT systems.

So that already kinda lets you kinda think about how how kind of small IT is on a on on a relative scale compared to everything else a company spends money on.

Which is usually headcount.

It's it's all headcount. And so, I mean, if you're obviously in an in an industrial company, you have a lot of cost of goods, but but a lot of, you know, meaningful portion is is headcount. So basically, if that's true, and you can make the labor side much more efficient, then could you pull off, again, a few more percent and and put that into AI? And that's the sort of size of the market. Now, importantly, what I said there was was a few percent, 5%, 10%, plus or minus.

This is not about all of a sudden AI getting all of the labor spend. I think this is much more of at Box, as an example, would we be fine to spend 5% of an engineer's salary on the equivalent of AI? Absolutely, because it's gonna make that engineer two times more productive, three times more productive. That probably won't impact our rate of hiring of engineers. That will just actually impact the amount of software that we ship.

And I think that most companies are gonna have that kind of approach where all of a sudden you'll just be doing way more as an organization.

Yeah, I mean, I think AI will replace some workers, but it's not going to replace all workers or all work. But you talked about this idea of new work. So do you mean categorically new work or just more throughput?

yeah, right now I'm gonna bet on just more throughput first, new work will come after, you know, again, that back to that idea of, like, right now, an agent is an extension of a person. I think it'll be a lot easier for us to contemplate the sort of more work scenario where, okay, previously I have a a lawyer that can review ten contracts per week, and now they can review twenty contracts per week. So that's not new work, but it's doing the work that previously we had this sort of arbitrary cutoff of.

And again, I think where people get this wrong is they think about sort of the demand of work as being relatively fixed, as sort of we we kinda magically have this equilibrium of supply and demand, and and we just landed on exactly what we need. But in our organization, for instance, we have a cutoff of what what sort of contracts we will spend time reviewing with our clients, or which ones will will we are we willing to kind of redline and go through because it's unaffordable for us to go below that line.

Now, in theory, you could have said, you could say, well, you could have just hired more lawyers and contract attorneys, and maybe somebody could have run that ROI analysis and it would have been worth it. But we have 900 other priorities that that we would be comparing that against, and so we just never would have gotten around to it.

Plus, then you unlock that bottleneck, but then you run into the next bottleneck, right?

Yeah, 100%. And so, but but actually, the it's it's good that you said that. So so what will happen is now with AI, the legal team will actually now go and experiment and lower that threshold or or do more than they would have before. And then this is the part that that also the economists always miss, is then what will happen is we'll actually be like, Wow, that was actually really efficient. We can now serve more customers than we could have before.

And then a new bottleneck will emerge after that process. And maybe that bottleneck, though, can't be automated. And so now we add more people in somewhere else in the in a workflow, because actually they you eventually still get constrained somewhere else in the system. And and these are the parts where anybody doing any economic predictions on AI, I just guarantee, or at least negative predictions on AI, I guarantee they miss that part because they can't model it, it's not possible to model.

These are dynamic systems that are too complex. You'll never model the unknown bottleneck of a company once they apply more automation. But what will happen is they will automate something, they will find a new bottleneck, they will have to hire people to solve that bottleneck or apply automation to it. But guess what? To apply automation to something, you need a person to make that decision and manage it for the foreseeable future. So that's where a lot of then the new jobs will come from that we can't yet sort of fully estimate today, but we're gonna see way more of that than than we realize.

Yeah, absolutely. I'm an early stage investor in AI startups, and you've said that now is the best time to start a company in the last fifteen years. Why so, and where do you think the opportunities are?

If you had to choose, like a few periods in history in the past forty years to be doing a technology company, I think you'd choose like ninety five to 2,000. Now, a lot of people freak out because they they would say, like, well, that was the bubble, but that also created Amazon, that also created Google. Like that was the first window where all bets were off, you could create a new company that that that could emerge.

Because you wanna look for these moments where something is being re-architected about the world. That's the only time that there's really a a new opportunity for a startup is something is changing, where there's sort of a need for a new thing that the incumbent solutions don't don't sort of solve for. The the most recent first window was right after the browser, and you could finally build internet websites that routed information or did e-commerce. Those were like the big categories, so you had two one trillion dollar companies emerge from that Google and Amazon.

Great, the next window was like, I'm gonna, I'm gonna sort of slip in Facebook into this one, but but really, the period kinda started like after Facebook. But there was like, oh, six to 2010, and this was like when we were figuring out that there was something about cloud and mobile and this new connectivity we have is sort of is allowing for new opportunities. So that created the SaaS boom, and it created the consumer web boom of Spotify, Instacart, DoorDash, YouTube, just the whole web 2.0.

But if you think about it after, like, 2010, eleven, twelve, there there was a little bit of a lull of those big brands. Like most big brands that we use today, that are these kind of next gen services, they kinda came out in a window. Because what we figured out was the template for how to do it. We we realized that our phones had GPS, we could order things, we could chat in new ways, the video was big, so that was kind of a period.

I'd say the past year and a half, two years feels the most like that second phase of any period I've ever seen, and probably on steroids. And this is a period where everybody's realizing that certainly in in some categories of software, there's not an incumbent that will be able to be competitive. And so you have a disruption opportunity in some categories of software because of AI. But the really big prize is that there's often not incumbent software in many of these new categories where agents are the software.

So there was never a an incumbent software vendor that did legal contract review. You had software that managed the data in a contract review process, but there was no software to do the review. There's no software that previously generated your code for you, right? Until you had agents, you couldn't do that. And so that that has created all of these new markets because software and agents can now go after non-software categories. So you you again back to that whole sort of like 10% of labor TAM, you know, whatever your number you choose, and you basically will either bet on companies that can do a very hard pivot in this direction with their current platform.

We're betting that we'll be one of those, and Salesforce is one, and ServiceNow is one, and Workday will be one for the its existing categories. But I would bet on startups all day long for all of the categories that didn't have a natural incumbent. And I'd bet on startups for a lot of these sort of overlays across multiple systems where no no system sort of owns that workflow end to end. So you have a new opportunity for these agents that might overlay multiple systems.

All of the rules are being rewritten, the the landscape is a is a full reset. It feels exactly like that sort of oh six, oh seven to to 2010, eleven, twelve period, where just it felt like it was complete open terrain on the internet and you could build anything. We have that right now with agents.

So exciting, I love it. Last question, and it's a question I ask of all my guests, What do you think it means to be human in the age of AI?

I'm, like, not philosophical enough to answer that question. So I I tend to not be too introspective, and, uh, I just don't think of AI in tension with with the human, uh, element. I I think it's a tool that will let us do way more. Certainly, in some categories, there's probably more of this existential question of, like, I trained up on this thing for many years, and now AI can do it. And I think that's a very serious not to be sort of just totally written off kind of element of of some of the transformation.

But by and large, I think humans want to be creative, I think they want to learn, I think they want to be social, I think they want to build things, I think they want to spend time with friends and family. And I think AI, to me, doesn't replace any of that. I think it augments most of that. I I would more bet on using AI to to do more things that you're passionate about and excited by.

And that's sort of how I think about it. I I would think about it as a tool to just doing way more than what you thought was possible before.

well, for not being very philosophical, I think that's an awesome answer. Thank you for joining us, Aaron. That was awesome.

Thank you. Thanks for having me.

AI is creating a shift. SaaS gave us software tools to do our work more efficiently, but AI will actually do the work for you. And Aaron saw that Box organizes your files better, your contracts, your invoices. But now with AI, he's building agents that review and approve the contracts that file the expenses. And this is what gets me excited. AI isn't just disrupting the software market, it's expanding the TAM by also disrupting the services industry and the labor market at large.

Thanks for listening, we'll be back next week. Pioneers of AI is a WaitWhat original production. Our executive producer is Eve Trow. Our producer is Rachel Ishikawa. Our senior talent executive is Stephanie Stern. Mixing and mastering by Brian Pute. Video editing by Eric Purcell. Original music by Ryan Holliday. Our head of podcasts is Lital Malad. You can join the conversation across social media platforms. Just look for us at Pioneers of AI. Thanks so much for listening.