The Exchanges, every show

Q I'm going to do a quick fire round with you. Uh, you have to go and be a public company CEO, I know. Um, so what have you changed your mind on in the last 12 months most significantly?

A I do think that, that I've, I've become more convinced that software is headless in the past year than I was maybe three years ago. And it's because of the, the level of agentic capabilities on tool calling and searching across systems and the accuracy of that. Uh, and that, that has happened faster than I, I would have, uh, perceived. So two to three years ago, if you were to kind of You know, wire up an agent and tell it, hey, go work inside a box and find a document to work with and do some process. It would, it would basically almost always find the wrong document and it wouldn't be able to handle actually like cracking open the file and reading through it. And so thus, you know, going headless wasn't sort of the, the most urgent priority, uh, from an agentic standpoint. And in the past year, those capabilities have just absolutely accelerated. To the point where I'm fully convinced that you just, you have to be, you know, headless first as a software platform.

AI assessment note: “I've become more convinced that software is headless in the past year”

Q Did you see the same thing happen with storage? I remember in the early days with Box and Dropbox and YouTube, you all had This major innovation with storage, and how did that play out?

A Let me give you a fun stat. We give our customers unlimited storage. We have 82% gross margin. So, so the, the, what, what happened was the price of the underlying storage has gone down by hundreds of times since we started the company. And then our, all our value is in the software layer on top of the storage. So we've benefited by this incredible, just, you know, ruthless competition between Western digital Seagate, other players that are just trying to pack more, more, more, you know, basically more, more, uh, storage density into these drives. And every couple of years they have a new break Through we're now, you know, upcoming, we're up, we're heading toward maybe a, a 50 terabyte hard drive. When we started the company, they were kind of 80, uh, 80 gigabytes.

AI assessment note: “what happened was the price of the underlying storage has gone down by hundreds”

Q Yeah. I mean, do you want to tell that story? I don't know if you've told the story. It's like a random intro, right?

A Well, it was just, he used to have house parties. Uh, TechCrunch had, had these house parties and And it was, um, probably no different than somebody's doing a house party in SF. Uh, you just go and you meet the VCs and founders and like, I'm going to make up examples. So I don't want to like, you know, there'd be like Chad Hurley over there pitching his, you know, YouTube to people. And like, oh, like that's just like how it worked. And it was just like, wow. Like that was this era where all these new companies were, were emerging. And I met our first investor, uh, in Silicon Valley at one of these house parties, Emily Melton, who then brought us into DFJ.

AI assessment note: “I met our first investor, uh, in Silicon Valley at one of these house parties”

Q Right. And so the week we're talking, you at Box are releasing a number of different agents. Um, let me start this discussion by just asking you, what is an agent? Because it does seem like it's an overused term and, and even myself who I'm, I'm in this all the time. I don't fully have clarity on what that word actually means.

A Um, I, I think the, ah, I think we should anticipate that it's fully overused. It, it is now the new term of art for talking to a, an AI system that is doing work for you. So just, we will hear, this will be the main term that we use going forward as an industry. And not because it's a buzzword, but actually it's a, it's a useful term. It's a, it's a definable object that is doing automated work for you. That could be in some cases as simple as answering a question. Um, but I think most people in, in the tech industry would generally argue that it should be doing some degree of, of work and looping through the AI model multiple times, um, uh, to do that work, and so, uh, that could be everything from, you know, very clearly something like Claude Code, or Cursor has an agent, or Replit has an agent, where you give it a task like, build me a website that has these qualities, and it will go off and do, you know, weeks worth of human work, In 10 minutes, and that's an agent that is managing that whole process, looping through the model multiple times, keeping track of what it's doing, updating its memory in the process, and that's effectively an agent. So that's an agent in coding, and we're going to see that same kind of agent architecture emerge in law, in healthcare, in finance, in education, where you can deploy agents to go off and do work for you. And, um, and, and there'll b…

AI assessment note: “a definable object that is doing automated work for you”

Q When you think about say like a box today, how do you think about leveraging LLMs? Is it a case of actually using six at the same time and switching between them for different use cases and purposes and being able to transition seamlessly between them? How do you think about that and the choice that companies have there?

A So we are building an architecture that lets our customers do what you just said. So effectively switch between models that they want to use for different purposes. Um, I think it's sort of less likely that we would be able to Kind of switch on their behalf in some kind of completely abstracted way because each model sort of, you know, is just a little bit different in, in how verbose it is, um, or how succinct it is, or what kind of style does it, does it sort of respond in? And so I don't think you're going to have complete commoditization of sort of the personality of the models for the, the sort of style and the response, uh, to the point where then, where then, you know, if you're a user, you know, any given response could come from Gemini, Versus GPT-IV versus, you know, Claude. I think that's probably less practical. Um, you're gonna be more wired into a particular model for some, some use case as a user of software. What we did basically is as, as soon as sort of this wave started, you know, let's say, 18 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, with, uh, with OpenAI's models. 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-IV is very good at legal, you know, answers, but Gemini i…

AI assessment note: “we are building an architecture that lets our customers do what you just said”

Q When you think about say like a box today, how do you think about leveraging LLMs? Is it a case of actually using six at the same time and switching between them for different use cases and purposes and being able to transition seamlessly between them? How do you think about that and the choice that companies have there?

A So we are building an architecture that lets our customers do what you just said. So effectively switch between models that they want to use for different purposes. Um, I think it's sort of less likely that we would be able to Kind of switch on their behalf in some kind of completely abstracted way because each model sort of, you know, is just a little bit different in, in how verbose it is, um, or how succinct it is, or what kind of style does it, does it sort of respond in? And so I don't think you're going to have complete commoditization of sort of the personality of the models for the, the sort of style and the response, uh, to the point where then, where then, you know, if you're a user, you know, any given response could come from Gemini, Versus GPT-IV versus, you know, Claude. I think that's probably less practical. Um, you're gonna be more wired into a particular model for some, some use case as a user of software. What we did basically is as, as soon as sort of this wave started, you know, let's say, 18 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, with, uh, with OpenAI's models. 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-IV is very good at legal, you know, answers, but Gemini i…

AI assessment note: “we are building an architecture that lets our customers do what you just said.”

Q Hey, Ern, a quick question for you. Obviously, there are a lot of software companies in the Valley. Based on what you've been through, if you were advising them on what metrics matter most, What's at the top of the list for them to pay attention to internally?

A The, there's probably a few that are just like literally the underlying business model economics. Um, I, I like, I'm a big believer in gross margin. Um, uh, your, your gross margin will ultimately determine your operating margin. Um, there's almost no way that, that you can, you can kind of make those two, um, you know, kind of get out of sync. Um, and so if you're, if you're subsidizing something or, you know, or, or in just such a commodity business and you're, you know, 40, 50, 60% gross margin, like there's just like, No way you're gonna have an operating margin that looks like a software company. Understanding your gross margin, managing to gross margin, I think is super important. Um, uh, you know, all forms of LTV CAC are, are probably good. I don't know what the latest, you know, make, you know, everybody, every two years is a new term in the industry that, that is used, but like something that just shows that you can acquire a customer profitably and whether the payback is a year or two years or three years, almost doesn't matter as much as, um, as, uh, do, do, you know, ultimately generate a, A long, a long-term, you know, sticky customer. Um, I think cashflow is, is super important. Um, I'm, I've, I've definitely like, I've, I've gotten religion on cashflow. Um, and I think companies, uh, getting the cashflow sooner, um, is a, is a really good move. I think it, it pu…

AI assessment note: “I'm a big believer in gross margin... all forms of LTV CAC... cashflow is, is super important”

Q years, not 20, but part of the challenge is obviously what are we going to do with this unstructured, slightly structured data, tagged, but massive, ingesting mass amounts of data that has a little bit of light structure to it, it has titles and search, and now you, it's remaking the company probably in some ways, right, where you can finally do more than text-based search across this data, right?

A Yeah, that's right. So we, we, we just, we, we lucked out. We have, we've been building a product for about three years. We just announced it called Box Hubs. And what, what Box Hubs does is, you know, traditionally you have your content organized in folders and you share those folders with people that you want to work with. Hubs creates a presentation layer where I can say, I want content from all these different folders to sort of be presented in this way to some audience. So Your sales team will have a sales portal. Your employees will have an HR portal. Your, your marketing, you know, all your employees will have access to brand assets via marketing portal. So that's hubs. We, we've been building that for three years and all of a sudden, you know, a year ago, you know, the LLM explosion happened and we said, wait a second, this is sort of the perfect scenario now where the, the, the customer is telling us what is their most authoritative set of content by topic. In a hub and you bring an LLM to that and now all of a sudden you can actually start to do more kind of question answer based, you know, like working through your data where you go to the sales portal and you say, what's the price of this product? And it's going to give you an answer instead of saying, here's the file that might contain the answer. So we think that's, that's a pretty big breakthrough in knowledge ma…

AI assessment note: “you bring an LLM to that and now all of a sudden you can actually start”

Q And do you think it's necessarily humans have a seat and agents have consumption, or would there be an argument for saying that, uh, agents in some way are not that dissimilar from, from humans, although they, they'll be doing a lot more with a lot more, uh, volume of data, and therefore there should be some kind of, like, seat-based pricing for agents?

A I think, I think this was, this is sort of a tougher category because, um, it all depends on the agentic use case. So, um, like I can totally see a world where we already have some customers playing around this idea of like, should agents have a box seat? Because, because why? Because they actually need to store data, uh, that gets retained and governed over a long period of time. And you want to be able to track it and manage it just like a person, but it's gotta be stateful. And so that kind of makes sense as like, we have to give it a name and a thing in our system to make that work. Do we charge the same as a regular end user seat? Probably not. Probably it's gotta be cheaper. Um, but then there's a lot of situations where the agent doesn't need an ongoing seat. They just need to be doing a lot of operations, in which case it's, it's probably just pure consumption. So I think it really depends on where does your software category land on? Is there a reason why you'd have an agent be stateful in that organization? Um, and, and, and kind of take on an identity and take on ongoing work. Uh, versus it's a thing that just every employee calls on demand. And that, that would probably determine, you know, what that business model looks like.

AI assessment note: “it all depends on the agentic use case.”

Q and then we did a pilot and that never really worked out. So now you're coming back to me like two or three years later and you say, no, no, no, no, no, no. Agent is the thing. And like this time, if you don't do it, you're going to, you're going to die. In the spectrum between skeptical and enthusiastic, where would you put the mood in large enterprises?

A I would say if I did like the broadest sample, I think the mood would veer statistically more optimistic than maybe the framing that I think maybe you landed on like a few extra cynical CIOs in that. I think because what's happening is the CIO and our main audience is the CIO. Uh, when we talk to the CIO, they know that their engineering teams are using cloud code and codex and cursor, and they're seeing the productivity gains come out of those teams. And they're like, yeah, like my teams are just building, you know, apps way faster. They're, they're being able to tackle IT projects much more quickly. Like we're, we're, we're doing security reviews faster. Like they're seeing the productivity gains in their, in their function. And I think they're often saying, well, how do I bring those same IT, you know, productivity gains To the non-IT parts of the organization, and they're having the business pull them and say, I want access to, to co-work. I want access to codex. I want access to these tools as well. So there's, there's actually a certain kind of sex appeal to, to these tools right now where the business is sort of demanding. I want to be on the agentic train because I'm seeing all these, these great use cases. And so I, I think the tone is actually remarkably optimistic and excited and positive as opposed to, you know, there, there's a sort of, You know, typical trough of …

AI assessment note: “I think the mood would veer statistically more optimistic”

Q IPOs and we've seen companies that are compounding faster than ever. Where do you think that leaves startups, including vertical startups? Where do you see the opportunities? Are we in a, in a world where everybody is ultimately either an open AI or an anthropic employee? Or, you know, service, uh, industry, uh, supporting them? Or is there room for lots of people to do, uh, lots of different things?

A I remain pretty, pretty confident and optimistic on the, the need for a kind of a bridge layer from the AI capability to the end user workflow. And, and some might sort of say that this gets kind of bitter lessened out, um, which is, which is, you know, oh, these things are wrappers on the model. And, and at some point there's a training run where it just like is the final training run That makes the, the renders the, the kind of vertical app or, or function specific, you know, app, uh, you know, not as useful. And, um, and I think that is a little bit too much of an accelerationist view of what people are doing with the tool, which is like, it's not just like what the model is spitting out or the model's ability to review information. It is how was the thing wired up into the business workflow? How did it get the context that it needed to be useful? I think if you're in a, in an, in an industry or a line of business There's a heavy amount of, of kind of integration with data sets, heavy amount of, of kind of bespoke workflows that that company does. That usually means that there's going to be a need for change management, implementation, ongoing support, ongoing expertise. And unless the labs build out literally the equivalent of hundreds or thousands of people for every single vertical and every single line of business, that means that there's actually a lot of opportunity in…

AI assessment note: “I remain pretty, pretty confident and optimistic on the, the need for a kind of a bridge layer”

Q Aaron, you've got to take attribution. As a venture investor, it's all about coining a term, ok? This was your original thought. In the shower, Aaron Levy's. Share it with me.

A I've been influenced by nothing I've seen online. Uh, this is all from me. Um, so there's some kind of, and who knows if this sustains as, as a full-time role or where it gets diffused into. I'm not, I'm not, uh, I'm not a hundred percent clear on that, but there is a hundred percent a role right now that there's going to be 500,000, a million jobs created for, and, and, and it's basically some kind of agent operator. And, and this person is, um, is actually going to be needing to be, uh, somewhat technical. They're going to have to like be deep in the AI world. They're going to have to understand MCPs and CLIs, and they're going to have to know how to write skills. They're going to have to understand agents.md files. The, the, it's going to be this group of people. That will know how to go into your marketing team or your legal team or your operations team or your life sciences research team. And this is the person that is basically going to enable that function to get leverage from agents. And, um, and the problem that the real world has that startups and, and frankly, many of your guests don't understand is that, is that when, when you start a company from scratch, you've got like, you know, the world is your oyster, right? You can design your workflows however you want. There's really no risk if you, if something goes wrong, cause you don't have much scale to begin with. Th…

AI assessment note: “it's basically some kind of agent operator”

Q But is this the same as the kind of classic VC blog post, which every fucking firm has written, which is like, You know, AI, it's moving from software budgets to labor budgets, and every partner goes and likes the tweet, and there's like, no fucking shit. Like, really?

A I mean, if you do it in that voice, it sounds, it sounds kind of like, um, you know, maybe, uh, you know, simple, but like, yeah, that, that, but like, that's just like a very big deal in technology. We've never had, there's never slash rarely been a technology that you could sell into an enterprise where you weren't capped by that company's Corporate IT budget. And so now for the first time ever, you have a technology where you can go into the line of business and you can say, I can now offer you a, a new tool in the form of an agent that will augment a workflow that will make you 50% or a hundred percent more productive. And so maybe I should be able to get five percent of your OPEX budget this year to go and do that. Like that, that is a new budget to tap into. And I don't think it like, You know, 10 X's the size of IT spend or technology spend globally, but it certainly doubles it.

AI assessment note: “yeah, that, that, but like, that's just like a very big deal in technology.”

Q Jensen very clearly said, AI won't kill software. It will explode the amount of software needed. And when I thought about that, you know, the thesis there is obviously kind of, you have this kind of core AI that crawls over 15 SaaS tools, and they really become databases that agents crawl on top of. Is that what it looks like? And are they not just valueless SaaS tools then?

A Yeah. I mean, I, I think that, that I'm, I'm sympathetic to that argument in some, in some categories. I think there's some software where, because the person was the user of the, of the software and they were clicking all the buttons that you're sort of ratio of buttons to underlying APIs was like more in favor of buttons. And I'm oversimplifying, but there, there are some tools where you open it up and there's like 93 features, um, that you're kind of clicking around on. And the, and the user has been so accustomed to exactly how to do that. That the, that the software's value proposition was correlated to, to roughly that, that sort of mass in a world of APIs and a world of agents being able to do more of the, of the work that you used to do on clicking those buttons. Then, then again, the value goes more to the API layer. So then the question is how many APIs do you have? Not, not in like a, you just need a thousand APIs, but like, like how robust and useful and proprietary and how much business logic is embedded in those APIs. Versus it's just calling a database and pulling a record. Like does the API surround a, a, a, a set of business logic of like, no, it actually secures the data or it knows exactly what person each piece of attribute should have access to inside the organization. That's, you know, at the end of the day, all software has a database behind it. So you co…

AI assessment note: “there's a lot of business logic in the layer above the database”

Q protect it. We mentioned our mutual love for Rory O'Driscoll. I do a show with Rory and Jason every week. Jason has bluntly said that this will be the golden age for cybersecurity because the security threats are going through the roof. Are you concerned with the system vulnerabilities and the security threats that are coming with AI? And what do we not know about security that we should know?

A I am concerned, but not, uh, in any kind of like new concern sense. Uh, this, this to me was kind of priced in the moment that we were generating code with AI. So if you can generate code, you have two problems. One, you're going to generate way more code than anybody's ability to review that code. So, you know, starting with GitHub co-pilot six years ago or whatever the date was five years ago, like that was just priced in, which is, which is as soon as As AI writes most of the code or, and then like 90% of the code and then 95% of the code, then by volume, we're just going to produce this unbelievable amount of code and, and any, you know, any change in a system, ah, you know, everybody kind of thinks about security as like, um, you know, is there a zero day where there was an unpatched, you know, component of your technology or somebody found a clever new packet, a package that, that, that you could kind of slip into, ah, every time you, you ship a new feature, You have a chance of a security vulnerability because the AI could have written in, oh, you know, we want to actually open up that port in, in the system because we need to do something. And maybe that was the wrong decision for the agent to go and do so. So we are going to be living in this new world of, of cyber risk, uh, in the form of, of using agents more. And then on the other side, obviously, if you have the of…

AI assessment note: “I am concerned, but not, uh, in any kind of like new concern sense.”

Q the one thing that I worry about is we see this insane demand side pull Every company in the world needs an AI story. Everyone wants to kick the tires with something, and I think we project the same demand side pull and extrapolate it continuously. Do you worry that we are in a momentary 18 month period on the demand side pull, and that may not always be lasting?

A You know, it's very possible I should be more sensitive to that. Um, uh, but, uh, I, I would take the opposite side of, of, of that particular wager at the moment because, um, partly because I already saw one diffusion cycle with cloud and actually how long that ended up taking and the, and, and the, the, the kind of spiky early nature, you would have just been like, oh my God, this is, this is on fire. It's, it's, and how could this last? And 20 years later, it lasted and got way bigger than we ever realized. If it works, the market's always larger than you ever think. And, um, and then the only, the only part why, why 18 months is like not even a relevant window to me is I think diffusion is going to take longer than Silicon Valley thinks. And it's back to the very first kind of that new role idea. When you go to most companies, they can't yet just deploy an agent to do, you know, full, uh, you know, financial proposals for all of their, their clients without a human reviewing the thing. And because the SEC will just show up and be like, Hey, like you, you just, you just gave this person bad financial advice and you're going to lose your license. Like that, that will just start to happen kind of across the board. And so, um, and, and so that that's why, you know, people take time. That's why we, we, we, there's a lot of regulatory controls and compliance teams, security teams…

AI assessment note: “I would take the opposite side of, of, of that particular wager”

Q It's, when we look back on this period and be like, what the fuck, companies trading at three times cash flow, like, way over-exaggerated or not?

A Well, three times cash flow is very much over-exaggerated. I would say that we're in a period right now where basically the market is being treated roughly as in, as the kind of indiscriminately, you know, kind of bucketed sector. And the next year, two years or whatnot, you'll start to see some separation and parsing between the companies, because as I noted in the beginning, agents will be really good for some parts of software. And agents will put pressure on other, other parts of software. So, and it'll mean some companies have to fully pivot and some companies can just sort of ride the wave. And if they respond, you know, effectively, like clearly three X free cashflow is, is like, you know, that, that seems like aggressively low territory. But I also think that at times in software, things have been aggressively overvalued, um, beyond the, the realm of, of likely what the terminal value is of, of, of a particular, you know, category or, or, or company as well. So I think we're, I think there's just a pendulum that, that needs to kind of find its equilibrium right now. Um, and, uh, and that, that'll play out over the next year.

AI assessment note: “three times cash flow is very much over-exaggerated.”

Q public sector, you saw a jump from 77 to 88% in accuracy for complex tasks. Healthcare saw a jump from 60 to 78%, and legal saw A jump from 57 to 69%, uh, uh, accuracy on complex tasks. Uh, that's pretty, pretty big. It seems like this model has, has almost been under hyped. Uh, can you talk a little bit about these, these jumps and what the significance is?

A I think, I think probably the, the main takeaway should be that, that the progress of these meaningful jumps that we've been seeing in AI coding Over the past couple of years where, you know, the model at best could do a couple lines of code You know, in a, in a kind of type ahead type format two, two and a half years ago in, in coding space. And now obviously people are giving the model a task of, you know, write me tens of thousands of lines of code for a full project. And, and we've just seen this incredible rate of progress and this March, uh, up toward, you know, more and more capability over time, uh, with, uh, within coding. I think that same trend is going to come to other Other now fields of knowledge work. And so, so this jump in sonnets model from four or five before six, I think represents an example of what happens when these models just get trained across more areas of knowledge work. What happens when they are getting better and better at reasoning capabilities that go beyond coding? What happens when they get better at using tools and deciding when to use tools? And that's what our complex work eval You know, is, is meant to represent is, is sort of how does it think through a problem? How does it decide it's got the right answer? How does it check its work? Um, and these models are getting much better at, at being able to deliver on that. So I think that'll be …

AI assessment note: “I think that same trend is going to come to other Other now fields of knowledge work.”

Q expert, but let's take it out a couple months, right? So you've got this data lock, right? But if my MCP is great, maybe I just grab those box files and that data, and I abstract away box. And sure, box has the data and the healthcare company. But if I can talk to it through a myriad of different levels and MCPs and everything, is it really a moat?

A I would actually argue that MCP is a thing that only enhances the underlying moat. Because data wants to be free. That doesn't mean that there's no moat around managing the data and making it useful. It wants to be incorporated in some other workflow. So for us, the more places where customers can use their data is only a net positive. We want to connect to every MCP, you know, environment that that's possible as a tool. And that just makes the continued feedback loop of having more data, at least in our environment, more valuable because, because you know that it's going to be accessible from all these other systems. So I think of that as kind of a, the analogy would be like, did APIs reduce people's modes? No, it actually in many cases made companies more sticky over time because they could be embedded into even more software, which made that even a greater level of, of sort of concentration. So, so I, I, I mean, I, I see no reason why MCPs wouldn't be just analogous to all of the, the APIs that we built over the past couple of decades.

AI assessment note: “I would actually argue that MCP is a thing that only enhances the underlying moat.”

Q Okay. Because your head's down executing or because you don't think it's a problem?

A I don't think it's a problem. Uh, I mean, probably if I'm like really selfish, like just means more shareholders for us. Um, like if you just on a continuum said, would you like there to be, you know, 10,000 public tech companies or 1000, like where would you want to be on that continuum? You know, you can, you can just instantly imagine the like liquidity dynamics that would emerge. Um, so, so I'm kind of neutral to like, I don't really care that much. Um, I think the, the interesting innovation, uh, that has emerged is, is, and this is, it's funny because people kind of, people kind of don't know exactly the, the root cause, like some, you know, maybe people think like, 20 years ago, we made it so hard to go public. I'm actually not in that camp. I don't think we made it too hard to go public. I think actually it's good to have a, of, of an, a heightened degree of scrutiny and, and regulatory pressure on being a public company. It, it, It is, is absolutely net positive for the average shareholder that we have all of these systems and, um, and governance controls in place. I think that's only a good thing. Um, but, but no matter how we got here, where we're at is we now have this new innovation, which is late stage capital that can basically keep you private for as far as we can tell, maybe forever, because you can, you can kind of outrun the problem of converting, um, these c…

AI assessment note: “I don't think it's a problem.”

Q Yeah. And we were talking about Snowflake a minute ago, and they did their, you know, Arctic model, and, you know, Databricks did, like, PJRX. Yeah. Did you ever consider building your own model, or is that just completely outside of the business?

A We considered it for 10 minutes, and, um, and we said, uh, no effing way, um, are we gonna get in this war? Um, uh, so it's, uh, you know, back to the Bill Joy thing, it's like, It's like, why would you want to, why would you even want the, the, the, the brain, you know, aneurysm of, of thinking about like, oh, do we have the, you know, all the latest researchers in our company and we're competing for that and we need to buy two billion dollars of compute for the next training run. Like that is just a war that you want to ride the tailwinds of that. Not, not playing that. Now, different question if we were Meta and Amazon and those guys. Um, but, but, but the moment you're not those, you have to like understand where you are in the ecosystem and then figure out, you know, how to ride that. So we even like, I, I, we, we debated for more than 10 minutes, fine tuning stuff. And even then we were like, no, cause like, you know, there's just model breakthroughs every day. Like what, why, why, why, why, why even have anything internally that could cause us to confuse ourselves That like our fine tune thing is like really important. You know, there, there's a lot of things that you can do where, where, um, and actually this was, this started to become an interview question that I had, uh, for people coming into the AI team. I was like, do you want to fine tune a model? And, and if the…

AI assessment note: “We considered it for 10 minutes, and, um, and we said, uh, no effing way”

Q Can I ask a stupid question now? Is that not what RPA is? Like, you know, we, we have large RPA provides. I always thought that was what RPA was.

A Everything I just said is exactly how you would have pitched RPA for the past decade. Um, and, and 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, it's often sort of, it's like looking at your computer screen and, and, and, you know, performing some kind of wrote, you know, routine actions doesn't handle variability very well. Um, because there's, there's again, doesn't have the, the level of intelligence that you now have in AI models. So, um, 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, but 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, you know, faster, 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. And, and it's kind of, it's a, it's a sort of real shift of how we think about software and, and the role of, of, you know, 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 ha…

AI assessment note: “Everything I just said is exactly how you would have pitched RPA”

Q I, you know, the show has been successful because I specialize in dumb questions from the way that you described agents. It was like the next generation of RPA. What happens to prior RPA providers? Does, does UI path just adopt an agent based model and actually move away from kind of rote learning? Or what happens?

A Yeah, I mean, actually, I think, I think being an incumbent RPA vendor is, is actually a great spot because, 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, um, and, 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, you know, kind of model for, for, uh, automation. Um, so I, I think, I think you could be very bullish on RPA vendors right now. Um, but I do think it means that more players kind of get in and around the space. So the, the, um, I think the, the, the thing that will, uh, 100% guaranteed happen, like, like in five years from now, this will be the most obvious thing of all time. But when you, 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 midsize or large enterprise or, or kind of developer oriented, you know, kind of individual. And so the, the total size of the market was basically arbitrarily or artificially held back by, by just the, the, the complexity of the legacy approach. So if AI makes it 10 times cheaper, faster, and easier to automate workflows, then, then it stands to reason that the market will be substantially larger. It could be a hundred, a hundred times larger At, at the en…

AI assessment note: “I think being an incumbent RPA vendor is, is actually a great spot”

Q On the AI agent space. And when we look forward, what do you think it looks like in five years time?

A I can sort of sufficiently say, hey, you know, 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 sort of, I think you'll have kind of category, you know, winners in, in a large portion of sort of job functions that today exist. And there'll be AI versions of those functions. Um, and, and this is a, this is again, one of these windows Where 10, 20, 30, 50 companies will get started that, that were like the window in two, in the mid 2000 where like all of today's SaaS companies basically emerged in like 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, you know, customer support agent, the AI, you know, marketer. And, um, you know, lots of companies won't work just for, for the, the, the, the kind of typical reasons. But we will have a landscape of, 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 all those different, that like AI labor? Like right now, you know, when you want to manage lots of software, you, you have, 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, I have digital labor, uh…

AI assessment note: “we will have a landscape of of basically labor that you can get from AI”

Q What did you not do with Box that you wish you'd done? You know, everyone has to prioritize. Everyone has strategic decisions on, we'd focus here. What decision did you decide not to do that you wish you had done?

A Earlier in our journey, I would have focused more on cash flow. I have become a little bit of a, of, of religious around cash flow. Um, uh, I think, I think owning your own destiny as a company is important. I think, um, you know, caring about every, and inspecting every single dollar of spend, In the business is very important. Um, I think, um, uh, I, I think these things, uh, you know, in, in, in sort of very loose capital environments, it, it sort of gets forgotten about or people don't really care about it. Um, but actually I think it helps you build a better business because you, you kind of apply constraints that force better decisions, better strategy, better execution. Um, so I would have done that, you know, years earlier than, uh, when we ultimately, uh, focus on cashflow.

AI assessment note: “Earlier in our journey, I would have focused more on cash flow.”

Q Do you think new juggernaut companies will be created both in the foundation model layer and in the application layer, or just in the application layer with incumbents gobbling up the foundation model layer?

A I would say that, that there will be some foundation model companies, um, but not, not nearly the, uh, the magnitude of, of the application layer companies. And, um, and much of that is, is due to the trends we've already seen, which is, you know, the moment you have, you know, people like Zuckerberg that are, are literally willing to spend billions and billions of dollars commoditizing the, the, the model layer, It becomes very hard to sort of figure out, well, how do you, how do you differentiate in that space enough where, where you won't, you know, kind of be taken out by, by one large training run from an open AI or a Google or, or a Zuck. There will be like niche or, or maybe, you know, industry specific, you know, sort of approaches you could take or some very, you know, 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, um, because, 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, uh, we, I think those, those will largely be subsumed by the, the, the big players with maybe room for one, two, three independent companies at scale that are not in the, in the hyperscalers, but there will not be room for 50 companies. Um, that, that's ju…

AI assessment note: “there will be some foundation model companies, um, but not, not nearly the”

Q Can I ask a stupid question now? Is that not what RPA is? Like, you know, we, we have large RPA provides. I always thought that was what RPA was.

A Everything I just said is exactly how you would have pitched RPA for the past decade. Um, and, and 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, it's often sort of, it's like looking at your computer screen and, and, and, you know, performing some kind of wrote, you know, routine actions doesn't handle variability very well. Um, because there's, there's again, doesn't have the, the level of intelligence that you now have in AI models. So, um, 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, but 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, you know, faster, 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. And, and it's kind of, it's a, it's a sort of real shift of how we think about software and, and the role of, of, you know, 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 ha…

AI assessment note: “Everything I just said is exactly how you would have pitched RPA for the past decade.”

Q On the AI agent space. And when we look forward, what do you think it looks like in five years time?

A I can sort of sufficiently say, hey, you know, 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 sort of, I think you'll have kind of category, you know, winners in, in a large portion of sort of job functions that today exist. And there'll be AI versions of those functions. Um, and, and this is a, this is again, one of these windows Where 10, 20, 30, 50 companies will get started that, that were like the window in two, in the mid 2000 where like all of today's SaaS companies basically emerged in like 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, you know, customer support agent, the AI, you know, marketer. And, um, you know, lots of companies won't work just for, for the, the, the, the kind of typical reasons. But we will have a landscape of, 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 all those different, that like AI labor? Like right now, you know, when you want to manage lots of software, you, you have, 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, I have digital labor, uh…

AI assessment note: “we will have a landscape of, of basically labor that you can get from AI”

Q I mean, also just for, you know, it's, uh, 10:30 here and I'm working. So, I mean, it's like, I literally had a reference call before this and they were like, dude, are you, Are you European? I'm like, yeah, I know, I know. I'm weird. Uh, but, uh, so are you nervous about regulation preventing progression in AI?

A I'm increasingly less nervous only because of what we've actually seen these bills sort of, um, come up with. They don't seem as, um, they don't, they don't seem as sort of progress halting as maybe what would have been rumored about a year ago. You know, the, the scariest moment to me was the pause AI kind of moment, which was, which was, okay, we need to, we need to stop the, the development of advanced AI for six months until we kind of, you know, figure something out. And, and it was like, it's been, let's say a decade of, of us all as a community talking about AI. If you think that in an extra six months is all it takes for us to have some kind of alignment on, on like, what, what is the doomsday AI going to look like? What is, What is sort of dangerous AI? What is less dangerous AI? Six months is not going to solve this. There are very fundamentally different philosophies in, in the land of AI that, that, that are irreconcilable. They will not, 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, you know, gum up the, the, the, the, the advancements if You know, government started looking at pause AI, and they're like, oh, even the tech community wants us to, to stop this thing. Um, and that, that was what I was …

AI assessment note: “I'm increasingly less nervous only because of what we've actually seen these bills”

Q So I actually tweeted that AI services companies are going to make more revenue Than, um, uh, any foundation model providers. And then Accenture posted 2.4 billion in revenue and OpenAI posted two billion in revenue. So absolutely right. Um, how do you think about that? Do you agree with me?

A Services are almost the leading indicator to compute, uh, because you need services to implement the thing that then eventually is sort of on autopilot, you know, not, not, not fully, but, but more or less. So, so I think I would take the bet on, on AI services for the next five years, unquestionably. Um, the amount of dollars that will go into the change management of systems, the implemented, the implementation of the technology is, is going to continue to be massive. On the other hand, the other thing that goes along with that though, is, is, you know, a bunch of the AI stack. So the actual GPUs, the, the data center build out that as well. So I think the, the, the, the, at some point though, the curve, if, if AI is, is as, You know, meaningful as, as I believe it is, and I think, you know, so much of tech believes it is. At some point, the, the actual AI, sort of, the software services of AI, this, 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 10 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, you know, five years ago when we were first moving more into the cloud, our services were…

AI assessment note: “I think I would take the bet on, on AI services for the next five years”