Box CEO Aaron Levie: CIO Advice on Agentic AI and the Enterprise

Yeah, welcome to 2026, the most important topic in the world. We get to live and breathe this every single day. As you noted, we get to work with nearly 70% of the Fortune 500. I think we're in this incredible moment of transformation, but we are still in the very early stages of what agentic work looks like in the enterprise and what the rollout looks like. I think we have an interesting dynamic, which is sort of a tale of two cities. We have AI agentic coding, which has clearly taken off. It's within engineering teams and everybody's kind of figured out what the new practices are around the future of engineering. Engineering is a complete vertical takeoff of AI agents. Then you get into the real messy environments of knowledge work, where things are quite a bit different. It's a lot harder to deploy agents at scale. The agents don't always have access to the right data. The users are less technical, so they don't know how to steer them properly. You have to do a lot of work to verify the work the agent is doing, whereas in software you can just test it and see if it works. You can't do that in a lot of areas of knowledge work. So we're in this really interesting phase where we've seen the promise of agents in coding, and the promise of chatbots in knowledge work. Now the question is what's the full promise of agents across knowledge work? This will be the defining topic over the next few years within the enterprise.

The promise of coding is honestly just an incredible gift. We've all done large engineering projects and we know how much time, maybe 60, 70, 80% of your engineering time, goes into the work that doesn't really make the project differentiated. It's necessary work, but it's the very labor-intensive, time-intensive things like upgrading a library version, doing edge case testing, building end-user features. That's where most of the time goes. A big project from a standing start could take six months or a year or two years before you see the value delivered. The promise of agentic coding is, what if we could have agents go and do lots of that blocking and tackling work? Then our job is making sure we're giving the agents the right plan, steering them in the right direction, reviewing their work, deploying testing and security solutions. Could we shrink that one-year project into two months? That was a fantastical concept people wouldn't have believed three or four years ago. Now that promise is reality. We're seeing this every single day internally at Box and from our customers. We might be able to literally do 3, 5, 10 times more work than we were able to do before, and the corollary is you might be able to do it 3, 5, or 10 times faster. At Box, it means we can build a much more significant product roadmap for our customers and take on way bigger problems than before. That's the promise.

Well, yeah. No, go ahead.

Yeah, so that's the promise and what we're starting to see the early signs of on the engineering side. Now everybody sits around and says, well, we want that exact same kind of output and outcome in knowledge work. You have a couple differences that have to be figured out. First, in engineering, the work is mostly in a text-based medium, writing code. AI agents have been trained on this data, lots and lots of code examples on the internet. The users are much more technical because they're engineers, so they can implement the technology themselves and keep up with all the updates. The work is more or less verifiable. If I build a bunch of code, I can do a regression test and see if I broke anything. Now compare that to knowledge work. You have a generally less technical audience. The work is less verifiable by definition. A contract, you can't compute whether it's correct or not. It has to experience reality, the red lines from the other party, or being taken to court. So you have to review the work a lot more. You have to review the final financial analysis to make sure it pulled all the right data from your spreadsheets. The agents don't have access to data in exactly the same way. In engineering, if you're an engineer working on a project, you already have access to the entire codebase relevant to that project. The agent also has access to all of that data. In the enterprise, we're constantly asking for permission to other systems, resources, and data environments. An agent is only as good as the data it has access to. We have lots of systems that are either not well-maintained, or the agent can't get access to the right data, or maybe even worse, you have too much access to information. We had security through obscurity in your organization, so now the agent is leaking data to the wrong people. Think about that T-chart I just went through and how many differences there are between agentic coding and agentic knowledge work. The project ahead for all of us across the economy is how do we take those same gains happening in coding and bring them to the rest of the world and our organizations? That is going to be a very big project that will take years and years of diffusion into organizations. I'm extremely excited by it because it means there are all new roles emerging. The upside when you're on the other end is tremendous because you can accelerate your organizational productivity. But make no mistake, this is a very real effort that companies have to go through right now.

Yeah, I actually didn't mean for this to become such a mini viral moment, but here we are. I have my own experience of AI and I've seen it from peers or other people in the industry. I believe I have a good pulse on this. I use AI as much as any CEO out there. I'm using it for prototyping new products, developing new ideas, doing lots of market research, automating customer intelligence. I'm using AI all day long. At the very start of that journey, you do these things like prototype a piece of software or do some market research and you think, oh my god, this thing is going to automate everything. You have this sort of existential dread for a few moments. For some people it lasts longer. I maybe got through it in a couple weeks. It ebbs and flows because a new model comes out and a new capability emerges and you think, oh my god, I'm experiencing it again. I'm about to get into the AI psychosis mode. There's a juncture where you can end up on one end, thinking we have to retool everything and what are the people going to do, and it gets very existential. The other end is you start to see that there are still lots of bugs the agent wrote, lots of security issues that were generated or discovered that now have to get fixed, and all the ongoing maintenance of the system. In the other areas of knowledge work, you find things like the agent pulled from the wrong piece of data, which meant if I hadn't reviewed that report, I would have come to exactly the wrong conclusion. I had to steer the agent a bit more, and it took a lot of work to prompt it in the right way. So on the other fork in the road, you end up post-psychosis, realizing that there's still a lot of work that has to get done to make these agents effective. It's a journey from instant existential dread to a bit more pragmatism. That's where I landed, and I think a number of my friends also landed here. This is a technology that is going to boost our productivity and let us do far more than we were able to do before, but it is not as doomsday as some of the commentary suggests. It's going to take a lot of time to fully deploy these agents to get the work done. That last mile thing, or the closeness to the problem, is really the issue. A CEO by definition is the furthest away from the real work happening in the company. You couldn't get further from it in any other role other than maybe the board of directors. So it's very easy for us to think we could just automate that engineer or that marketing campaign. But when you're closer to the problem, you realize you probably can't have an agent do all of that without any human supervision because it's going to do the wrong thing, or the taste of what the agent delivers is going to be off, or it's going to introduce a bug. That keeps the humans in the loop for the foreseeable future. The thing I encourage CEOs is to use the technology so much that you get to the other end of that psychosis and you can actually see in a much more practical and pragmatic way all the places where humans are still necessary to get the ultimate gains from this technology. Don't just stop when you prompt an agent and ask it to generate a research report and get blown away, thinking that's the only thing that's going to happen in the future.

Oh man, this is a fun one. I think it's a fantastic question. My instinct is always biased toward being more ambitious with what you can do. That will more often lead you to seeing better productivity gains, but also see the edge of this technology much faster. Anytime I hear stories about people not getting real ROI from agents, I think it often approximates not pushing them hard enough and doing very menial type of work, as opposed to pushing the limits of this technology and getting the gains out of it. It's hard to have a full general piece of feedback other than I would always bias toward being more ambitious, pushing the limits much further. Importantly, if something fails with AI, with whatever that ambitious idea was, you almost have to try it again six months later, almost every single time no matter what it is, because the model progress that's happening right now is still at an incredible rate. You often have to say, forget about the fact that six months ago this didn't work in this particular part of our codebase or marketing campaign. If you relied on your understanding of image generation as a marketer from six months ago, that's already been blown up. The latest image gen models are perfect at being able to do text and have photorealistic capabilities in a way that wasn't possible six or twelve months ago. If you're in software projects, something like Fable or GPT 5.5 is another step function improvement on what these agents were capable of. We had a project just yesterday where we tested Fable versus Opus 4.8. You'd think they should be on the same trajectory of progress, but the thing that Opus 4.8 couldn't do, Fable just finally did. If you don't constantly recheck what is now possible with AI, you will fall into the trap of believing that the productivity gains aren't there or you're not able to accomplish as much as you wanted. So have very ambitious expectations, constantly be trying the technology, and keep pushing the limits.

Yeah, we've drunk the Kool-Aid on this one. I think there's probably two versions of it, so it's hard to know exactly which one the question is coming from. I actually think there's going to be a role for effectively an internal FDE. This is some kind of IT business AI automation engineer type role. I think often it will live within the technology or IT organization but be embedded within the actual line of business it's trying to bring automation to. There's the internal version and then there's the external version coming from either the vendor, the systems integrator, or maybe a new consulting firm. If we go back to what's the difference between coding agents and the rest of knowledge work, a lot of it does come down to how technical the user is and their ability to navigate around all the challenges agents run into. One of the big challenges is, do agents have access to the right data to work with? Do they have access to the right internal systems to pull context into the context window to be effective? An engineer knows their way around those different systems. They know how to get access to resources, set up MCP servers, set up CLIs, and deal with compute sandboxes the agent might need. This is just not how most knowledge workers were trained. We're used to going into PowerPoint, Word, Google Docs, Slack, generating stuff, creating files and sharing them, and doing lightweight manipulation of these tools. But for an agent to be really effective, they need access to all the right data. The data has got to be in the right format. You probably need skills files and readme files in the process. This is the agents.md phenomenon that has emerged. You have to have the permissions to turn on MCP servers. All of that is technology work. There's almost no way for that to not be done by somebody who is either formally trained or has figured out how to be highly technical. The job of FDEs is to go help companies understand today's workflow and then figure out how to bridge the capability of the technology that is emerging and constantly changing with the workflows of that organization. Sometimes you can get lucky and embed agents in an existing business process and it miraculously works. But often you have to re-engineer the process, migrate data sources into a modern system, and change the workflow so agents can be more effective. All of that is effectively the work of the forward deployed engineer. That FDE could come from a vendor like Box, Salesforce, or Palantir, or from a systems integrator or a new consulting firm. I think there are going to be all these new roles that emerge, which are basically the technical talent that has enough business acumen to get into the business processes of that organization and successfully implement the tools of AI to bring agents to bear in those workflows.

Yeah. Unfortunately, that question is a maybe priceless, unanswerable question in any kind of binary sense. I think the answer is I'd rather bet on both. It kind of depends on which dimension you take that from, but I think we're still at a point where I would bet on either a technology company or an end user of the technology that has simultaneously the frontier models they're leveraging and the best ability to get those models the right context to work with. Sometimes you want to do this thought experiment: if you were starting your company from scratch, what would it look like to get the full gains of AI? It's pretty intuitive that if you could start your company from scratch, you'd basically design your business processes in such a way where the agent has an innate ability to get access to the context it needs to help you automate work. If we started a one-person company, you'd probably start with a file system that had all the context of your business, all of your goals, your marketing strategy, your customer language. Then you'd build your company up from that foundation, knowing agents always need access to the right data to answer the right question or automate the workflow. So you would do both: the frontier intelligence and you'd have proprietary information and insight that you're giving that agent, and you'd develop a flywheel. As you hired more people, you would try to maintain that ability for agents to keep having access to the right data. You'd really value sources of truth and authoritative knowledge banks. One of the problems traditional enterprises have is that their sources of truth are everywhere and many are not well-maintained. If you were starting from scratch, you would always write down your final decisions in a way that agents can get access to, so the agent doesn't land on the wrong resource or get the wrong insight. So to the question of what you should do right now in any meaningful or large enterprise, you are basically trying to do whatever you can to get to the point where a company from scratch is built up in this kind of way. I would be betting on frontier intelligence right now with the combination of access to the right context for those agents to be effective, and making sure your organization is built in such a way where you can get those agents the right data in those workflows. Now to your second question about commoditization, architecturally, does that mean you bet entirely on one frontier model? Probably not. You probably want to design some kind of neutrality. Right now, you should still try to exploit the gains from frontier intelligence as much as possible, but in one, two, or three years from now, you're going to see a stratification between the cost of frontier intelligence and the cost of the second-best frontier intelligence that can do the job extremely well for a bunch of long-tail tasks. You're going to end up in a world where model routing becomes very important. Maybe something like a Fable-esque model gets the planning part of the work and the review part, but the in-between massive token usage comes from a more cost-effective model for that type of work. So you want to invest in the right kind of architecture that can deliver on that type of outcome. That's probably where we're heading from a tokconomics standpoint. Making sure you can design an IT architecture and a data architecture that lets you deliver that is super important.

Yeah, I think I thought this was true in the kind of cloud and SaaS era, but now it's an order of magnitude more true. We used to have this thought process: if you show me your IT stack, I'll show you your culture. I could tell from the tools you use the kind of culture you have, how fast-paced your company can be, how innovative it is, because you can see if the average employee has access to the data they need to do their work and can communicate instantaneously with colleagues with very few barriers. That was the version from 10-15 years ago when we saw the rise of Zoom and Slack. I think there's a new version today: if you show me your IT stack, I'll show you what you're going to be able to get from agents. You can just see instantly what kind of agentic productivity and outcomes you will get based on where your data is today, how much is in legacy systems, how much is in systems where the agent has access to these tools, and what frontier intelligence you're using. Oftentimes we're having an AI conversation, but it's really an IT architecture conversation. The companies that are going to win are the ones that have the most modern, well-maintained, well-structured data environments and IT stacks that allow agents to be effective. That is the single biggest determinant of how much value you're going to get from AI.

But it's sort of masking an IT architecture conversation or a data architecture conversation. And that's really a lot of the conversation we should be having, which is: do you have the right data platforms? Do you have the right IT architecture to let you get the real productivity gains from AI right now?

Yeah. Well, we definitely preferred the world where it was being subsidized. So those days are over. It was fun while we got it, guys. There was a brief moment where you could rely on the subsidization from venture capitalists and then get tokens for your coding agents.

It was a beautiful moment in history. It lasted about a year, and we all had a great time. There was definitely a period where you could have really exploited it and had 10 years of software development paid for by LPs and VCs. Unfortunately, those days appear to be coming to an end. But now we're in a much healthier environment where you're more or less going to be paying for the real underlying costs to deliver this. And then it's just capitalism, microeconomics 101: where are you getting the right ROI from these agents? You should deploy agents where it's most effective, where you are getting that ROI. We're fortunate to be in a bit of a sweet spot because, as an enterprise software company, we have a good pulse on our customer base. So we know how to deploy agents in a way that makes us more productive and lets us ship more software that we think is valuable. We're smack in the middle: if you had a token-maxing company, we're not that—we don't have a leaderboard internally or incentivize the most tokens. On the other end, a company where everything is locked down and hard to access frontier agents—we don't want to be that either. We rely on more mature, sophisticated approaches: as long as we have a really good product roadmap and good product managers and engineers working on designing that, then I'd rather move that product roadmap forward as fast as humanly possible. If token spend goes up exponentially, that should be correlated with a good thing: we can deliver more software to our customers faster. Then it's more of a CFO exercise: how do you plan for that? How do you budget? What are the trade-offs? I do think the corollary is we're well past the point where the IT organization can hold the entire budget of AI, and that's actually healthy for this technology. IT budgets run at maybe 2-5% of total revenue depending on the industry, and that's an artificial cap on AI's potential. Maybe you want AI alone to be 5% of your total revenue, so that would double the IT budget. To do that, you need the line of business to own the budget and the deployment of AI in the organization. So we're in a renaissance of the IT organization: your job is to bring intelligence to the entire business and be the experts in the technology, partnering with the business on where it gets deployed.

Yeah. I think this is an incredible time if you're super curious, if you're excited by technology, if you're one of those people who geeks out on the latest iOS release or now the latest Claude Code or Codex release. So I would go very deep in the technology, experimenting with it constantly to push the limits of what these tools can do. The awesome thing is that it's one of the first moments in history where almost everybody has the same amount of access to information at the same moment. Twenty years ago, there was a huge advantage if you were in Silicon Valley—you'd see the iPhone first because you could go to the developer conference—but today in 2026, maybe I have an hour heads-up of information from anybody else because of some chat thread among Silicon Valley founders, but otherwise we all have access to the same technology at the same moment. That means if you're paying attention to the right resources—this podcast being one of them, there are other great AI podcasts—you can be as informed as the best expert in the world. You get the same news feed as Andrej Karpathy, Greg Brockman, Dario Amodei, and Sam Altman. So there's very little they have access to that you can't take advantage of. We have this cool moment where information sources keep up with the pace of innovation—our organizations don't, but our ability to tap into what's happening does. So I would just be playing with this technology constantly, pushing the limits, breaking things, and trying to figure out the implications for an organization I already work with or want to work in. Those kinds of people will be the ones that get ahead.

Yeah, no, it's a key insight that some people miss. You're so used to thinking about this as anthropomorphizing the technology—maybe too much—but it's not just an engineer or a marketer. It has all of the skills in one. I often go to AI for product prototyping and give it a very specific prompt, but sometimes I'll add 'feel free to add anything else you come up with' into the prompt. It will come up with an idea that far exceeds what I ever would have been able to prompt because it can take in all these extra domains it knows about. So when prototyping, try not to limit yourself to your own imagination. Give it that extra nudge to come up with better ideas. The cool thing about AI—getting more expensive, no question—is you can have five tabs running and have it do five different versions. I often prototype in parallel to see if there are things I'm missing and expand the use case or capability I was trying to come up with.

Yeah, there could be a few different ways to take that. I think this is a big question across the software ecosystem: what's the new value proposition of software in a world of agents? I think you're not going to go and vibe-code a CRM system or an ERP system. You shouldn't be vibe-coding a core system of record for your documents and enterprise data because you're benefiting from 100,000 other customers that need the same technology. It's the vendor's job to get really good at that one thing. With AI, we can do that faster and better. So there's been confusion about where we'll apply these tokens and capabilities. The value of core systems of record continues to matter a ton. But then the question is what can you do with agents on top of those systems? That's what gets me excited because we're going to use our tools far more. For example, I use Salesforce more today than at any point—maybe five times more—because I have an MCP server connected to Claude Code. I'm constantly using that server for customer and market intelligence that I'd never do manually. So I think there's been a misread on the market opportunity: we're moving to a world with maybe 100 times more agents than people using software. What will we use these tools for? So many more things. We see this from customers processing data at a scale never possible before. All of a sudden, the value of unstructured data assets, CRM systems, and ERP systems has gone up. That's how value shifts. Governance becomes really important to ensure you get those gains.

Sorry. Okay, my bad.

Yeah, I think maybe I'll start generically and then give examples. You should really treat this as a technology for abundance. You almost have to look at your business and ask yourself: if I had unlimited capacity in a certain area, what would I do differently? Unlimited capacity for combing through information, using judgment, accessing data—what would change about that workflow? You quickly realize there are many parts of your business constrained by the ability to deploy human resources. If agents can do that work, what would you do differently? For example, as a B2B company, if I could deploy agents to comb through our customer base, we'd have better insights on when to reach out. If I could deploy unlimited compute, it would know everything about every customer so we can be more helpful. In HR, I posed this to our recruiting team: if you could comb through LinkedIn and then hop over to the internet to see GitHub projects, thought leadership in the press, and build a full profile, and at the right moment when something's going on in that organization talk to them—how would that change recruiting? So you can go through your entire organization and see where your business could have different returns with unlimited capacity. To a CFO, there are huge examples: what if I could analyze my business totally differently, see where there's waste, and reapply dollars to growth; which customer segments are unprofitable; where are market opportunities? All these are constrained by the number of people working through spreadsheets, ERP systems, analytics data. Now you can throw compute at that problem and you're no longer constrained by team size. So we're in a renaissance of workflows we can execute, but that takes looking across the organization for where you'd have more upside with unbounded resources.

Yeah, definitely a big fan of value maxing. I'm glad that terminology has already taken over from token maxing. Honestly, it's hard to be generalizable. It's probably no different than asking 20 years ago, before any AI or cloud technology, how to measure ROI in marketing, sales, or finance. You'd have to use the tools we've always thought about: I have a certain amount of dollars to deploy against various things—events, marketing campaigns, products, hiring. How have we ever measured ROI? Among all these choices and opportunity costs, where is the most effective use of the incremental dollar? The bittersweet news is it's no harder or easier with AI than at any other point. It's always been this squishy thing with judgment where sometimes you guess wrong and sometimes right and you double down. AI is no different. You need high-judgment people. The one difference: with the wrong prompt and limits, you could spend $50,000 overnight, so you have to be thoughtful about where to deploy tokens. But short of catastrophic mistakes, the problem looks similar to business history: you need people close to the business with budgets and high judgment, who understand the technology and what it's capable of, and a rigorous process to review ROI from deployments. This is not a one-shot environment; it's an ongoing budget management process.

I do think it's really important to keep up with the technology. I don't know of a replacement to that, unfortunately. I know that was the qualifier to the question, but you just have to find a way to stay as current as possible, and it behooves companies to educate their employees on this.

I think that's like 90% accurate. The person who owns the delivery of a project is in the best position to know the productivity rate they can get with AI, assuming they have religion on this. The exception is when things are so expensive or transformational that you need senior leadership to get behind. But generally, I'd bias toward that point.

Yeah. I think the combination of technical skill and business acumen is still the best approach. If you're interested in marketing, get really good at marketing and the core principles, but also understand how AI accelerates building a campaign or doing market research. Living at that intersection is the most potent way to deploy this technology. So if you're doing clinical drug trials in life sciences, deeply understand that but also understand how something like Claude Code or Codex accelerates that kind of work.

Squishy for sure. I think it's mostly embedded, but there are some areas where it has to be an adjunct.

Yeah.

I feel the pain all day long, so I completely agree.

I think this is the most exciting moment in history to be a CIO. There's so much change in technology. If you're even remotely curious as an IT leader, there's an incredible tsunami of new things to play with and try out. The importance of the CIO role rises dramatically because agents are perhaps the most technical solution ever deployed to non-technical people. You're deploying non-deterministic intelligence into the hands of every knowledge worker—it could run wild, grab the wrong data, produce the wrong report or the right one. That's all determined by your technology architecture, how you've deployed agents, and how you've trained users. So the CIO's role becomes substantially more important. You're effectively providing work to your organization—the first time in history where IT is responsible for deploying the actual output of the organization, not just the tools that enable it. So it's an incredible time to be in IT.

Because of our customer base and how we've grown as a platform, we see the danger in too much information in the hands of the wrong people or agents. You have to make sure only the right context gets to the agent. You can't just shove as much context as possible—first, it's a security nightmare; second, it'll actually get the wrong answer because it has too many things to pay attention to. So the right context at the right time with the right guardrails is still critical for agents to work.

Our job is to help customers solve many of these problems with their unstructured data. The reason we're so excited and passionate about AI is it's the first time in history we can tap into all this data in our organizations. At Box, we're building the leading AI platform to help companies tap into unstructured data and knowledge and work with your entire agentic ecosystem. I appreciate you having me on; these are the topics we get excited by.

Thank you.