Aaron Levie, CEO of Box, on Box AI, Enterprise Enthusiasm, and the Evolution of SaaS

The rate of change that we're seeing and the rate of just exponential improvement that we're seeing from AI models is incredible. And I would assume if we keep that pace up, I think these systems will increasingly be able to perform any kind of general task. Jensen Huang at NVIDIA kind of put it the best, which is, you know, effectively the IT department becomes the HR department of AI. And that just opens up so many new questions about what the future of IT looks like. I think we're entering a new era with systems of intelligence that let us combine data, AI, and underlying enterprise software to go and automate really anything about our business. There's going to be a tremendous amount of AI startup opportunity, but it will not come from just doing an AI-first CRM system because you should anticipate that Salesforce is an AI-first CRM system.

Hello, and welcome back to The Cognitive Revolution. Today, I'm excited to share my conversation with Aaron Levie, founder and CEO of Box, the intelligent content cloud. Box powers secure collaboration, manages over 100 billion documents, and is now introducing AI-powered workflow automation for more than 100,000 customers globally, including such diverse household names as pharmaceutical giant AstraZeneca, legendary nonprofit Teach for America, investment bank Morgan Stanley, peer-to-peer rental market Airbnb, the $1 trillion chipmaker Broadcom, and even the United States Air Force. As you'll hear, Aaron says that leadership at these organizations is now more energized about AI than they've ever been about any other technology in the history of his career, including cloud computing, where Box was an early enterprise SaaS pioneer. While cloud adoption faced initial skepticism and resistance around security and compliance and just needing to do things differently, Enterprises today are eagerly exploring use cases for AI. In many cases, Aaron says, perhaps more than is immediately practical. Box is, of course, racing to meet the moment by building new AI functionality into all aspects of their platform. And they recently launched Box AI to help enterprises get more value from their unstructured data in a secure, controlled way. We discuss several key capabilities, including natural language querying across enterprise content, automated metadata extraction, and their vision for AI agents that can autonomously perform workflows like contract review and routing. Aaron shares a number of technical details on how Box is now layering foundation model capabilities onto their existing foundational features, like their Hubs product, which they originally developed before the current AI moment to allow users to curate canonical authoritative versions of key documents, and which has proven to be the perfect base on which to build accurate, reliable retrieval-augmented generation experiences at scale. We also dig into key questions facing enterprises and the software companies that serve them, including how IT departments will need to evolve from supporting work to actually performing work with AI agents, the transition from per-seat to consumption-based and other pricing models, and whether startups can compete with incumbents who are racing to add AI capabilities. Aaron makes the case that while incumbents will continue to dominate many established markets, There is still massive opportunity for AI-native startups to build fundamentally new products that work across platforms and address unmet needs. As always, if you're finding value in the show, we'd appreciate it if you'd take a moment to share it with friends, leave us a review on Apple or Spotify, or drop a comment on YouTube. And always feel free to share your feedback either via our website, cognitiverevolution.ai, or by DMing me on your favorite social network. I got a lot of encouraging messages after the recent AMA and R1 episodes, and it was great to hear from so many listeners. So please do keep it coming. For now, I hope you enjoy this conversation about AI adoption in the enterprise and the ongoing transformation of enterprise SaaS with Aaron Levie, CEO of Box.

Aaron Levie, founder and CEO of Box. Welcome to The Cognitive Revolution.

Thank you.

Yeah, good to be here. A lot of, a lot of stuff going on in AI land right now. How you doing?

Never a dull moment, that is for sure. Well, let's do a quick warmup and then we'll get into what's going on with AI in the enterprise and then, you know, what you guys are bringing to the enterprise with your latest AI features. For starters, what are you doing with AI in your personal life? And what is your AI worldview, particularly as it relates to, like, are we going to get AGI soon? Or what are you sort of expecting over the next couple of years?

Yeah, well, I think hard to obviously say based on the very amorphous definition of AGI. But so I'm kind of like everybody else downstream of whatever Ilya or Sam or Greg are talking about. So I have no better ability to predict where that's going than I think anybody else right now. But All I would say is the rate of change that we're seeing and the rate of just exponential improvement that we're seeing from AI models is incredible. And I would assume if we keep that pace up, I think these systems will increasingly be able to perform any kind of general task that you give it. And with these reasoning models, I think we're seeing incredible results in math and, you know, very complex logic and obviously coding. So once you have those basic foundations, you know, if we could continue to improve more and more on the benchmarks, that gives you the building blocks for basically models that can continue to learn on their own and get, you know, as advanced as we need them for basically any task that you'd want to give them. So it seems like within the next few years, we could be getting some form of whatever we would have previously defined AGI as.

Yeah, not that much farther to go, I would say. And certainly the checkpoints seem to be coming, if anything, more densely than they were not long ago.

Yeah, it's great.

How about in just like your day-to-day use? Do you have kind of favorite use cases? Do you feel like you're properly challenging 01 Pro or are you doing more basic stuff? Where are you at?

I think, you know, partly because I always get enamored with new technology, so I try everything out. But I would say one thing I was actually reflecting on recently is I have more new apps on my home screen in the past, let's say 6 months than probably any other time in the past decade, decade and a half. My home screen was like, okay, you added Uber and then you added Spotify and you added, you know, maybe one social app and then WhatsApp. And like only every one or two years did something get to the homepage. But recently I have at least 5 new apps that I've added into the mix, which to me is a little bit of a proxy for just how much infusion of AI has already occurred in our personal lives. I'm, you know, talking to Gemini Voice or OpenAI, you know, with ChatGPT with video. Pretty regularly using Perplexity to get different kinds of answers, playing with xAI and Grok. And it's just incredibly exciting to see the range of what these AI models can do. Built a prototype with Artifact and Claude. You know, everywhere in my personal life and kind of professional life, AI is kind of getting added into the mix.

So recently you put out a LinkedIn post that I thought was pretty interesting where you basically said, that this is the most energized you've seen enterprise companies about a new technology at any point in your career.

Yeah.

Obviously, you made your company on the cloud wave. So this is, that's another big wave that people were pretty enthused about. How would you, for people that aren't in the room and aren't hearing these conversations with senior leadership at enterprise companies, what is the vibe like? What are they doing? Are they hands-on?

What are you seeing?

Yeah, you know, given that you brought up cloud, it's actually a really interesting comparison between the two. So in the early days of the cloud, I actually would not describe the energy as super excited, super animated, and aggressively, you know, pursuing moving to the cloud. I would say most of the conversations we had with enterprises were there was some degree of skepticism, there was resistance, there was a lot of friction. There wasn't a lot of just pure excitement with no friction around it. The reasons for that were, this is a very big shift for enterprises. They had to move their infrastructure from the data centers that they managed to the cloud. They had to trust these new vendors that they hadn't worked with before. If you were an enterprise, you'd never worked with Amazon, you know, as an enterprise vendor. That was obviously very atypical. And then you had to sort of change your whole notion about data privacy and compliance from a world of of sort of physical management of hardware to a world of, I'm going to get a set of APIs and maybe some dashboards and some audit reports, and that's my only way of kind of controlling these systems. So enterprises were very reluctant early on to move to the cloud. If I compare that to today with AI, recently I was in New York, met a couple dozen CIOs and customers, and the reaction was, if I snap the line at like 2 or 3 years in the cloud versus 2 or 3 years into AI, You know, couldn't be more different in terms of the environment. Enterprises are looking for almost as many use cases as possible that they can deploy AI in, probably in many cases more than is actually practical. You have a sense of creativity and excitement and innovation that didn't necessarily exist in the cloud. You know, with the cloud, it was kind of like very pragmatic. It was like, you know, I could take this piece of infrastructure and move it into a virtualized environment. There's nothing like that exciting about that. It's kind of very utilitarian. You know, with AI, people are saying, well, what if we could actually solve this business problem that we've never been able to attack before? Or what if I could go and deploy human resources at much more interesting problems than where my talent is currently, you know, dedicated to? And, and all of a sudden, the, the creativity, the energy, the opportunity is totally different for these enterprises. So there's a lot of excitement. There's a big kind of asterisk, which is we're still very early. The actual amount of at-scale deployments in the enterprise are still very early innings in terms of where has this technology been deployed, but the excitement level is totally different. And so, what this means to me in being in enterprise software is there's just going to be an insane amount of opportunity. We are going to see opportunity for bigger companies like the Microsoft or Oracles or Googles of the world. There's going to be Opportunity for, you know, the software stack like Box or Salesforce or ServiceNow. There's gonna be a lot of opportunity for brand new startups to go and build AI agents that solve a whole new set of problems for the enterprise. And I think there's a very, you know, it's probably a relatively narrow window, but there's a window right now of opportunity where you're gonna see a tremendous amount of change, a tremendous amount of new startups that get to emerge. And enterprises are very much, you know, ready and excited for pursuing that.

What do you think that change looks like in practice? One of the comments that you made in that post was that you kind of expect IT departments— in a way, this could almost be sort of a reverse from the cloud, because in the cloud it was like, now you guys aren't going to manage this physical stuff anymore, and there's going to be more kind of put over on the vendor side in terms of making sure we have uptime and whatever, and we'll just kind of consume that and build apps on top of it. But if I understand you correctly, you're saying almost the opposite when you say IT departments are going to go from supporting the work to actually doing more of the work. And I wonder, like, what does that mean? And how many enterprise IT departments are up to that challenge today? Or how much are they going to have to transform to take on that challenge?

I think enterprise IT departments are going to have to transform pretty dramatically. I'd say that You know, history of IT was many times in partnership and even proactively with the business. The IT department would work with the business. The business would say, and by the business, I mean the marketing team, the sales team, the finance team. You know, somebody in the business would have a particular need. You know, I want to have a CRM tool so I can track all my leads. I want an HR system so as we hire more people globally, we can make sure that we're managing all the local laws from an HR standpoint. They go to the IT team. They review a set of vendors, and then basically it gets handed off to IT to go manage the system, deploy the technology, pick the vendor, and then enable the business. And what, you know, that relationship has always been, you know, has been well-defined and sort of codified for, you know, a few decades as we've had kind of modern IT environments. But it was still the responsibility of the business to make use of the technology and to be productive and to drive execution in the company. So, You know, the sales team still was owning all of the execution and they were using a CRM system from the IT team to be productive. AI kind of flips that on its head, which is now it's not just the people in the business that are using the technology as sort of using it to enable the business, but the business is going to go to the IT and say, I actually need you to go and deploy AI labor against different kinds of problems that I'm dealing with. And so now you might have a future where The head of sales will go to the IT team and say, "I need to spin up a new sales campaign. Do you have AI agents effectively that can go and help me do that?" Or a back office processing, you know, team might say they might go to IT team and say, "Hey, do you have AI agents that can go review all my contracts or review all these invoices or help with all my client onboarding workflows?" So this is a totally different era for IT to be in the position of. Now actually solving the business problem, not just deploying the software to help the business solve their problem. And this means you're gonna have to understand the business way more if you're in IT. You're gonna have to become even more strategic for the company. You're certainly gonna have to understand all the trends happening in AI. You're gonna have to sort of understand the ecosystem and all the surrounding, you know, companies that, that produce AI agents and all the software around that. you know, Jensen at NVIDIA kind of put it the best, which is, you know, effectively the IT department becomes the HR department of AI. And that just opens up so many new questions about what the future of IT looks like, all of which are much more exciting, I think, than the past. But we are in for quite a bit of change in this space.

Yeah, gotcha. Okay. So if I torture a metaphor here, if the IT departments are the HR departments of AI, Then Box is applying for or looking for promotion in a lot of these organizations and has just been through what you might call an upskilling with a bunch of new generative AI product and feature releases. So tell us what's new as we enter 2025 at Box.

Yeah.

So for us, what we do is we work with about 115,000 customer companies that are our customers. And what they use Box for is they use Box to secure, to collaborate, to automate workflows around their content. So, that could be their contracts, their financial documents, their marketing assets. It could be their research data, their product plans, all of that, those documents, those media assets, the digital images, all that content we store, we help secure, and we automate the workflows around or allow, enable companies to collaborate on that data. AI is really the next frontier of now what we can do with all of this unstructured data in the enterprise. And so we built a layer called Box AI that connects AI models to enterprise content in a secure way that lets the enterprise continue to have their security, their privacy, their access controls, their permissions. So we handle all of that, and then we connect to AI models with this abstraction layer. So what we've been building out is a set of capabilities to let people really transform and unleash the, the power of their data. So the first big use case is, you know, being able to talk to your data for the first time. We store well over 100 billion files in Box, and you can imagine every single one of those has incredible insights and value inside of it. But most of the time, you don't know what's inside of your data because until you search for it and look at it, you don't know what's actually out there. So now, for the first time, you can just talk to your data. I could take 100 sales presentations and say, I need What's the best practice of me selling this product to a new customer? And it's going to read through, our system will read through all those documents and then produce exactly the right answer for whatever you're doing with that customer. Or you could take a bunch of medical research or life sciences documents and research and be able to talk to all of that data to find interesting trends or patterns in that information. So that first use case is basically, you know, retrieval augmented generation on many documents with an AI layer that kind of powers that. The next big use case, which we've just been starting out with, is the ability to read through any kind of content and then extract the most important structured data from that content. So take a contract and you want to be able to pull out, you know, the renewal date of the contract or the party names from the contract. And then you want to store that in a structured database so that way you can query it later. You can have dashboards, you can automate workflows. And so that's the next big use case that we're working on, which was, is sort of rolling out imminently. And then where this is all sort of going is eventually agentic workflows within our platform or connected up to other systems. So I think the, you know, the very big prize in this space is, is if you take all of the information in your enterprise and what if you had agents that could go off and operate on that data? And so I could have a contract agent or a marketing agent or a sales agent, and it could perform tasks for me that make me more productive. So Go review the contract, pull out the riskiest clauses, go and route it to the right person. And I could basically design an agent to go and do those workflows for my entire dataset and then connect it to another system. So I want to connect it to Salesforce, I want to connect it to ServiceNow, I want to connect it to Microsoft. And Box becomes really this layer for managing intelligent content management, but then we connect to other technologies that that will handle the AI in their particular part of the workflow. So this is the future that we're sort of building out. And I think it's the biggest set of change and the most amount that we've been building in the history of the company. And it's unbelievably exciting because you're on a daily basis seeing new things that were never possible before with our data.

Hey, we'll continue our interview in a moment after a word from our sponsors.

2 kind of digging in questions on the retrieval and then the agents one by one. So retrieval, I think obviously RAG has been a huge trend and exactly what you described. A lot of people and companies have rushed out and implemented a version of it. And then I've heard over and over again that it sort of hasn't worked super well for people. Usually when I get under the hood and kind of explore, like, why isn't this giving people the answer that they want? It seems like the most common failure is that the usually vector embedding mediated search is not retrieving the right content in the first place. And so you don't have the right sources, then you're not getting the right answers. What have you guys done to deal with that? Perhaps the structuring of the unstructured is a part of it, but how do you— address that problem, which seems to pop up. It's kind of the first trough of disillusionment for so many.

Yeah, 100%. Now let me ask a question. When you see that trough of disillusionment, is it often with datasets that are quite broad or heterogeneous? So we've seen this a lot with the use case of, I have my email, I have my data, my calendar, and I want to do a search across all of it with some kind of AI system. Are you seeing it in those kind of scenarios? Or, or other ones?

I feel like almost always, kind of honestly, even if it's not like a crazy complicated situation, it still feels like it pops up.

So I, I'd love to attribute this to pure genius. We got lucky. We were building a product about a year before ChatGPT launched, which was, it's now called Hubs. And what it was, was the ability to organize content on a topic by topic basis. So you could share that content or search that content on a per-topic basis. And with this sort of novel idea, which was within your Box account, you manage files inside of folders, but you could create a hub that basically points to content within your Box account. And it was the whole power of it was to create a sort of a many-to-many relationship. So I could have the same document show up in 20 different hubs without ever moving it, without ever changing the permissions on it. And so if I made one update to that document, it then propagates to all the hubs that it's accessed in. So we created this architecture because what we saw was a lot of people wanted to create a sales hub where it pointed to the sales data, but then they said, I want a sales hub for the Japan region, and I want to create a sales hub for a particular product line, and I don't want to go and change the documents and sort of have to fork them and keep track of why is the one in the Japan hub updated, but the one in the product hub is not. So the whole idea was create kind of virtual pointers back into the same source of truth of data, but then you have all these virtual hubs that you can create. So we were working on that about a year before ChatGPT, unrelated to AI, but with sort of search and discovery as one of the common use cases. And then as soon as ChatGPT launched, we were like, oh, like obviously the next big thing that we should do is let you talk to all the data in a hub. And what we discovered was This was sort of the breakthrough architecture for a RAG use case because where RAG runs into problems is let's just pretend you had 100 million files inside your enterprise and somebody went in and they said, I want to find what was the last revenue figure in our earnings, our last quarter earnings. And the challenge with RAG is that you might have a document called like Earnings_final and earnings_draft1 and earnings_draft1_sally edits. And the challenge is like the RAG, I mean, the vector embeddings on all of those documents look very, very relevant to being able to produce the answer. But the accuracy and authoritativeness of any one of those versions might be totally wrong based on where it was in that editing process of that document. Now multiply that by years of data and signal from other systems like email or other data sources. All of a sudden it becomes very hard to then be able to go and ask any kind of generic question on your internal data in the enterprise. The reason why this is less of a problem for the public services like Perplexity or other is because you get the benefit of almost like a PageRank algorithm for public data where you can sort of look at a, you know, a sort of a curve and say, you know, This CNN article is more authoritative than this random blog, or this blog is more authoritative than this very random website that was created 2 days ago. And so you get a little bit of an authoritative score in the public internet in a way that our corporate data doesn't really have. Our corporate data is much more messy. It tends to not have a kind of a PageRank element to it. So it's very hard to know the ultimate source of truth. So hubs basically solve this problem because what happens with hubs is users are basically telling us What is the authoritative copy of the data that I'm putting into a hub? So if I created a sales hub that has the sales presentations and product information, I'm only going to put the ones that are the authoritative records into that hub. And then when you're going and doing a question in that hub, you're only going to really be asking it questions about sales. You're not going to ask the sales hub an HR question. You're going to go to the HR hub. And so we've been able to kind of get the user to ask the right types of questions for where the dataset is. The dataset is authoritative, and it's not as messy as doing a broad-based RAG environment on all your data. So those are kind of like 2 or 3 different hacks that we've kind of lucked into and then kind of doubled down on that has made our particular RAG service, I think, you know, at least 100 times better than just doing like broad-based deployment across all your data.

Yeah, that's interesting. I've heard so many of these stories now about how Somebody just happened to be sort of building the right thing that was, you know, perfectly complemented by AI. And, you know, then the whole business shifted as a result of those things coming together.

We pinch ourselves because if we hadn't been building that, I do kind of like, I do get very scared because it was about a year, year and a half of just deep architecture work. Like there was no way to build it any faster. Like you had to create a file system that could have a virtual sort of sharing component. And, and for the complexity of our platform, that just took a year plus. If we had not already been a year into it or a year and a half into it, I fear that we would have felt it would be too daunting based on how fast AI was moving that I'm not sure we would have landed on this exact architecture. We may have done, you know, kind of like a less optimal version. So in this case, we got totally lucky, but ultimately built the exact right thing for the exact right moment.

Yeah. Okay, that's cool. On the agents one, obviously, 2025 is the year of agents. We thought it might be 2024. It turns out it's 2025. What does an agent mean to you? I have my sort of sense of what the definition is, and it kind of— I contrast an agent with what I tend to call an intelligent workflow by how much autonomy or decision-making discretion Yeah. The AI ultimately has. How do you think about it? And, you know, how much autonomy or discretion are you giving to the agents in your platform? Yeah.

Yeah. Yeah. So I like that definition and I would fully subscribe to that. We, we have in our, in our platform, we, we've taken probably the broadest definition only because what we didn't want was there to be like, you know, you, you hate to use software where there's like 17 different versions of a thing and then the use, you're, you're kind of like, pushing all that complexity on the user to understand the differences. So what we, we've defined agent as basically a mix of an AI model or many AI models, a set of tool use within the platform, and then effectively underlying skills or capabilities that it has, which is some mix of like, you know, system prompts, but also some kind of proprietary architecture change. And then of course, access to your data. So that, that whole collection of capabilities creates an agent. So then within our platform, you can do very simple things with an agent. You can go and talk to a single document and you're talk— you're using it, you're talking to an agent, but it's not doing anything agentic. It's just you're talking to the agent that is the either the Box, the core Box AI agent, or you're talking to— you could create a custom agent. So you can create a sales agent that has a custom prompt and custom instructions that just have you know, information about, about how your sales workflows work or the kind of language that, that you should use. And so you can create custom agents to let you talk to your data, to let you extract metadata from your documents, to let you talk to lots of files at once, to create content. So that's our first era of agents. I think many of these agents would be what we would've called assistants 2 years ago. But again, we've created just a universal language for this. Where it's really going is much closer to your definition, which would be agentic workflows where the agent is performing many sequential tasks and there's some degree of probabilistic elements to those. So that would be, I want to review a document and then based on how I reviewed that document, I want to kick off another process to either another agent or a human. I want to be able to take a lot of data, collate it, and produce something. These are much more multi-step flows that we expect you will make very agentic in the future. And so right now, you know, software, the limitation of software is that software is very good at these deterministic workflows, but the vast majority of work is actually not deterministic. It's probabilistic. It does require judgment and you do change your answer based on other inputs or insights. And so the majority of work in the future will be non-deterministic. Judgment-oriented, agentic work, which is an incredible opportunity for enterprise software because that is what we can now finally go and digitize. So the way that I've been thinking about this recently is if you think about the eras of enterprise software, 40 years ago, we had the initial wave of systems of record, which was CRM systems, ERP systems. This was the definition of the most deterministic technology possible. Like you're changing rows inside of databases effectively. So, so that, and, and, you know, based on that, it kicks off a thing and that was like 100% deterministic. Then we moved to systems of engagement, which was this idea of like collaborative systems, Slack and Box and other tools where things are a lot less structured. They're a lot less deterministic. The workflows are a lot more fluid. They can adapt a bit more because it's kind of human to human and a little bit messy. And then with this new era that we're in, more, more of systems of intelligence, this is really the era of now the AIs are automating those workflows and they have all of the properties and benefits of, of being structured like systems of record, but they have all of the flexibility of a system of engagement where it can adapt, it can change. And that, and that's because that's what now AI can do. I think we're entering a new era with, with systems of intelligence that let us combine data, AI, and underlying enterprise software to go and automate, you know, really anything about our business.

Hey, we'll continue our interview in a moment after a word from our sponsors.

Yeah, I often say intelligence, of course, this has been debated for a long time and I don't pretend to be the final word, but my working definition of intelligence is the ability to do useful work when there is no explicit algorithm that tells you what every step ought to be.

Yeah, I think that's a great, very practical definition, especially for the enterprise use cases where you just don't want to kind of have a predetermined or pre-wired workflow because it's going to change. You need it to adapt to new data. Maybe there's not even APIs available for that thing you're trying to do. And that, that's where even things like browser access gets very exciting because there's a lot of future workflows where I want to go and automate these 5 systems and how they talk to each other, but there's not clean APIs to have them all go talk to each other.

Yeah, I find myself in my AI-assisted coding workflows When it's working well, I'm basically copying and pasting stuff around most of the time.

Right.

And when it's not working well, then I start to have to troubleshoot. But, you know, the smooth thing is like, I'm just kind of the copy and paste monkey that's gluing these other things together. It's a weird experience when it's working well. So what would you say are the biggest bottlenecks that enterprise companies face today In terms of realizing the value? I have a thesis that like the AIs are actually good enough to do a lot more than people are in fact deploying them to do. Tyler Cowen recently said, you know, in front of an audience, you, all of you, the humans, you are the bottlenecks. What do you see as being like the practical barriers that people are currently still struggling to get over?

Yeah, so I 100% agree that AI can do vastly more than what most enterprises either think or have even the near-term, you know, kind of appetite to go and deploy. So there's to some extent, you know, more technology available at the moment than most people realize. And even myself, I have to like remember, hey, that task that I normally would have gone and pinged somebody, hey, can you go work on this thing for me? I'm reminding myself more and more, no, like go and try and create it in Artifact or use AI to go and do this. And so even right in the center of AI, watching everything happen around the world, I have to trigger my brain to remember. You know, actually how much these things can do. So I can only imagine if you're not in it every single day, well, that gap is is probably fairly massive. That being said, I don't want to let the AI model providers off the hook though, because one of the the things that does prevent AI deployment is you know you can't have an enterprise workflow in a particularly a regulated industry that works 98% of the time. You would not find it acceptable if 98% of your flights. That you scheduled were successful and 2% of the time you show up at the airport and you don't actually have a ticket, right? And that's enterprises need 99.99999% reliability on almost anything that's really important. So if it's a creative task, you know, write a marketing campaign or review a blog, write a blog post or send an email on this topic, you have some kind of room for a little bit of hallucination or you can go in and have the ability to go and edit it later. But if you're going to have a billion financial transactions get submitted every week or every month, and AI gets 0.01% of those wrong, then that's a non-starter for deploying any of these kinds of systems. So I do think that what we need from the model providers is, and you can see it in the benchmarks, we're getting closer and closer, but we need to be getting to even higher degrees of accuracy. We need to get until like all of the evals have to be completely reset. You know, so that way, because everything has hit 100%, I think we still are a little bit technology dependent, a little bit AI model dependent. We do need the costs to continue to come down in AI because this is actually due to the excitement from customers is customers will say, hey, I'd love to go deploy 10,000 AI agents at X problem. But then they look at the cost and they say, okay, you know, I still can't yet afford what that would look like in my business. Or maybe we have to take it in a more stepwise fashion until costs come down. So I think we need AI to still be cheaper. We need to be able to get the performance of these models up even higher. And then from there, I think you're, you're running into, and to Tyler's point, all of the classic human-based change management difficulties. And those range from still privacy concerns and security concerns in some cases. Those range on, you know, the very real issue of, hey, I still, I actually have humans doing that thing. And so until I can kind of transition them to a different role or teach them a new skill. You know, we're not going to be able to go and automate that particular workflow. And so all of that is, is still what we're in for in the enterprise. And that's going to take years and years. I mean, this is going to be at a minimum, a decade-long change for, for how enterprises go and become more AI-first. But I think we have a roadmap because we did it in cloud. And I think we have an increasingly clear vision because we sort of start to understand what this world could look like as, you know, what would an AI-first enterprise look like? How would agents be a part of the workforce? How do we now get better insights from all of our data? How do we automate almost any workflow in our enterprise? And I think that you see this increasing understanding of what an AI-first future could look like in most organizations.

Just to push a little bit harder on this, I feel like, you know, there's sort of a corollary to my definition of intelligence. If intelligence is something that allows you to do work when there's no explicit algorithm, A corollary is you should only use intelligence where there is no explicit algorithm, right? Like, if you have a way to do something with traditional code, you probably should. It'll be faster, cheaper, more reliable.

Yeah.

So if we then bring the AI back into the domain of all these fuzzy things that we don't have these algorithms for, and we ask, how good is AI at doing these things as compared to a human? My belief is that with some elbow grease of making sure you have the right context and putting a few examples together and whatever, applying all the best practices, I would say you can most of the time get to the point where the AI can do as well as the human at much reduced cost and much faster. And we've seen that for no less than medical diagnosis recently, right?

Yeah.

So if you buy that, doesn't it sort of suggest there's still something else going on? Like, Maybe it's a fallacy, or not necessarily a fallacy, but sort of an attitude of like, we need the AIs to be not just on par, but actually have a 10 times lower defect rate or something. I mean, that seems to be the case with self-driving.

Well, I actually think you just nailed it. And this would be like advice to any software vendor. You can't go to a company and say, I can do exactly what you're doing today and you're going to save 40%. You know, like, in like, and this is like an economist would say, oh my God, everybody would do that deal all day long. But, but like, once, once that meets real life, that person has 17 other projects. There's an incredible amount of people and attention and priorities that are all competing for their time. So if you could wave a magic wand and make something 40% cheaper, like, you'd totally do it. But like, of all of the things that I have to do, Like that just might be like number 9 on my list. And to your point on that, is there something else going on? Like that person now has to go to the AI council and get approval and they have to go and prove, they have to do a full 6-month test to make sure that it actually is 40% cheaper and all of that work. And then they have to weigh that against anything else that they're doing in their business. So to your point, I think we need AI to be producing multiples better improvement on the status quo. That's how you really compel motivation. You can't be incrementally better, incrementally cheaper, incrementally faster. You have to be an order of magnitude better on one of those dimensions. So if you could go to a company and you could say, listen, for— I can, I can be literally a 10th of the cost of, you know, what, what you do today to go review your contracts or review your invoices or automate this backend supply chain process, then you're talking. Then, then you're like, okay, I could save you millions of dollars. Or if you could say, I can actually do a 10x better job than your human-based workflow today. You know, we'll discover that cancer more effectively, or we will actually be able to target even more automation across your enterprise. This is often why actually I think AI's biggest opportunities— I think that's the flavor of what you're saying— is that to actually go after the work today that is not automated. So you're not even replacing something that's existing, you're just layering onto an existing workflow and just making it better. I, I would say this largely explains the breakthrough in Copilot or Cursor is because I get to do exactly the same thing I'm doing now, but I'm seeing incremental productivity gains right away without changing really any behavior. The more that AI can solve those types of problems where it's all a net new use case that just adds incremental increased productivity, easier on the change management front. Anything where you're replacing an existing process and you're only saving a little bit of money, Way harder problem to go after in the enterprise.

So what do you make of the debates around the future of enterprise software? I mean, we've heard— I feel like people are groping around, of course, because everything's happening fast and we're trying to make sense of it. But I feel like we've heard pretty clearly conflicting narratives. On the one hand, it was like, oh my God, now it's never been a better time to start a startup. And then it was like, well, actually, though, this technology is so easy to implement that probably incumbents will capture most of the value because they'll be able to roll. You might have a little bit of a go-to-market lead, but they already have the customers, so they'll beat you if you're trying to start up versus have them just figure it out and deploy to their current customer base. But then we also have the Klarna narrative where they've allegedly reportedly shut down a couple of systems of record. And then there's also the pricing debate too, as to like, you know, do we still charge by seats anymore? Or do we have to go to like per outcome based pricing? So there's a lot there that could probably consume the rest of our time. What do you make of all that?

Exactly, exactly, man. If I had heard that when I was just starting Box, I would be way too stressed to start a company. So there's definitely a lot of variables that are in flux. I compare that to our early days, which was, it was, you know, lots of other problems, but not like the fundamental variables of the business model of the company. And I think you're right. I mean, you have a sort of which spaces are incumbents, the natural, you know, sort of have the natural advantage, the underlying billing model of this. Is it seat-based or is it outcome-based? The idea of, is there a future of some, you know, AGI light that makes some software just irrelevant? You don't even need the software in the first place. So all of those things I think will happen. I would say, you know, I would more lean toward timeless lessons. of competitive strategy, which is, you know, if you're a brand new startup, you go after things that are not easy for the incumbent to go after. So if all you were doing was building a sort of a thin layer on top of OpenAI, bad idea. If you're building a thin layer on top of Salesforce with AI, bad idea. So, you know, Salesforce is very competent. They will build the CRM AI thing. Workday will build the HR AI thing, you know, ServiceNow will build the ServiceNow AI thing. And equally, if you're just finding a slight gap in what OpenAI does today, you'd have some risk of them moving up the stack or the model getting better and better and kind of, you know, eventually bringing that into, into the model layer. That being said, I can think of a number of things that are, would be unnatural for OpenAI to do. Because maybe you have a lot of non-AI interface and workflow work that has to happen to make X problem get solved. Or you can be doing things inside of the sales world, HR world, or this ITSM world that the incumbents equally aren't going to do. Maybe it's cross-platform AI workflows that are not natural for, you know, any one of those players to sort of lean into. Maybe it's building AI agents that are so so kind of like orthogonal to the normal strategy of Salesforce or Workday or ServiceNow that they, that those companies wouldn't kind of think to pursue them. And you can get enough traction fast enough where you have some degree of a moat. So I think there's going to be a tremendous amount of AI startup opportunity, but it will not come from just doing an AI-first CRM system because you should anticipate that Salesforce is an AI-first CRM system. And I think we're seeing startups all the time that are finding those windows of opportunity right now.

So what do you make of like a Klarna? I mean, I've heard every take on that from like, it's all hype. They're not even really doing it to, yeah, maybe, but that's the exception that proves the rule.

I think right now it's the exception. I think I've sort of, as I've seen the reports, I'm more in the camp of maybe it's overplayed a little bit, but nothing about it is impossible. So given that it's not impossible to do what they've said, then maybe they have just chosen that this is gonna— this is, you know, something that will make them differentiated as a company. That's still different from every company on the planet, you know, doing it in a homegrown way. I mean, my understanding was they were going to build their own Workday system with AI, and that's just like not a priority. Again, back to that prior point of like, where are you in the prioritization stack? Most companies are just not focused on building their own HR system to save a couple hundred thousand dollars. So I think what they're doing is super provocative, super interesting, and yet not translatable to the broader economy. But it's super fun to watch. I invite as many companies to try that experience and share their lessons along the way. I think it makes the conversation in the ecosystem much more interesting and much more, much more dynamic. But I'm not convinced that 90% of corporations would ever do what they're doing.

Any pricing guidance?

Yeah, this one's, I mean, I don't really have any guidance because we ourselves are going through the testing of all the business models. In general, I think one of the biggest benefits of AI is that you can get much closer to software solving an outcome. The conclusion of that theory would be that then you should have your pricing model be closer to that outcome. That could be either very literally the outcome, i.e., You pay an AI agent to generate leads and thus you pay by lead. Or it could be, you know, what is the consumption that goes into that outcome? I want the AI to generate 10,000 leads. And so that takes a certain amount of compute capacity. And so I'm paying for the consumption of that compute capacity. And then there's just like, you know, traditional subscription models, which is, okay, I want an ongoing license. It'll roughly do this much volume for me. Sometimes it's a little bit more expensive. Sometimes it's a little bit more volume. Sometimes it's a little bit less, but I pay the same. Fixed rate the entire time. I think we're going to see every version of these business models get pursued. And again, I kind of put it in the category of super intellectually interesting because we're in such a dynamic period. We have not had, you know, open questions about business models in software for, for again, 2 decades. Like, you know, CRM— Salesforce basically invented, you know, maybe with a couple other companies, basically invented the idea that you pay per seat on a subscription basis. And that's been the business model of SaaS for 20+ years. And now we have a chance to say, oh, there's other business models that will begin to emerge. I find that incredibly fascinating, but I think each company has to kind of pursue its own understanding of what is its customer looking to pay for. And so then thus which business model kind of makes the most sense.

One niche one that's really interesting to me is in terms of things that people might want to buy separately. You know, if they're like going to continue to be a Salesforce customer and get all the agents or whatever, but maybe one thing they might want to buy separately would be sort of safety or compliance kind of stuff. Somebody to like red team, for example, the agents. Do you see that as viable?

Yes, absolutely.

Okay, cool. That's interesting because the other option would be like Salesforce would just deliver that too as another feature.

But like that. So it's really good to understand like IT stacks. And so if you just think about an IT stack of, okay, I've worked at Salesforce, I have HR, I have an ERP system, I have Box, you know, which things cut across all of those? And that's where you need a new vendor that is sort of independent of any one of those players. And so some things lean toward, well, actually it really only makes sense to be in one of those systems. And then other things lean more toward, well, no, I want a technology that works across all of the apps in my enterprise. And then that kind of determines whether you could be a startup or whether it's really an incumbent game in that, in that market.

That's, that's good perspective. Last one. Of course, it's been famously said that we see the impact of computers everywhere, but the productivity statistics, it seems like AI is maybe still in that zone, but I wonder what you are seeing internally at Box when it comes to AI-enabled productivity boosts. And is that something you can measure? Is it something you're believing in and encouraging on faith right now? And what do you have for other leaders who want to make sure they're getting the productivity that's promised?

Yeah, so we are 100% committed to being an AI-first enterprise and company. And then that's fairly important for us because we sell AI technology to enterprises. So we need to be the first to understand where this is all going. But also, I just think it's going to be a way you run a better company in the future. And so that has a handful of ways it's showing up already. So one, with Box AI, this is, you know, sort of how we work with any of our unstructured data. So if you're a new employee and you want to learn how to sell our product, you go to our sales hub and you can ask it any kind of question and then it'll give you an answer back. And it's basically like you're talking to a top expert in the company. So you are getting all of the value of talking to You know, the smartest existing employee, but now you can do it 24/7. You don't have to wait for somebody to respond to your Slack message. That's on one hand. That's sort of the part that's probably less measurable because it permeates everything we do and just improves productivity. And then in other areas where, again, it's kind of anecdotal, but we've deployed AI coding tools and I get ranges of, you know, somebody will say, hey, it was 5 or 10% more productive. Maybe a new hire, it's 50% more productive because they can just ramp up so much faster. I think this will be, you know, most, it will show up in the biggest way as we will just ship more software. That will be the measure of productivity that we care about internally. But I think AI will certainly be the first kind of discrete technology category that shows up as in that kind of long-term GDP graph that people tend to talk about as driving productivity gains due to technology.

So over half a percent a year, which is what Tyler Cowen recently said, you take the over, it sounds like.

If you can let us start the clock in a few years, I think you still have a diffusion of the technology across the economy that takes way longer than I ever think it should. But that's just life in human-based, you know, human-mediated environments.

Yeah, I'm with you on that. I always underestimate the timelines.

Yeah, exactly.

Anything else you want to touch on or leave with the audience before we break?

Yeah, I mean, I think we're just in such an unbelievable time to be building technology, deploying technology, you know, that like, The fact that anybody anywhere in the world could be getting like 90th percentile expertise on any topic instantly is just an unbelievable time to be alive. If you had even contemplated that 3 years ago, based on our understanding 3 years ago of where technology was, it would not have been conceivable. So, I just think it's an incredibly exciting moment and we're super excited to bring it to the enterprise.

Yeah. Terrific stuff in both respects. Aaron Levy, founder and CEO at Box, thank you for being part of the Cognitive Revolution.

It is both energizing and enlightening to hear why people listen and learn what they value about the show. So please don't hesitate to reach out via email at tcr@turpentine.co, or you can DM me on the social media platform of your choice.