Okta's Oktane 2025 | Ben Kus, Box

Clips
More from Okta's Oktane 2025

Ben Kus

Chief Technology Officer

Box

play_circle_outline Overview of Box's focus on unstructured data management and collaboration.

play_circle_outline Case studies showcasing AI's effectiveness in retrieving information for sales teams.

play_circle_outline Enhancing AI Security: The Crucial Partnership Between Box and Okta for Safe Data Access and Interaction

play_circle_outline Call for enterprises to leverage trusted platforms for AI capabilities.

Info
Transcript

Ben Kus, Box

Ben Kus

Chief Technology Officer Box

At Okta’s Oktane event, theCUBE unpacks exactly how Box Inc. is unlocking unstructured data with AI that thinks, reads and reasons. Rebecca Knight, host, and analyst Jackie McGuire sit down with Ben Kus, chief technology officer of Box, to unpack why 90% of enterprise data requires smarter tools. AI classification to secure sensitive content and how agentic AI retrieves precise answers across massive file sets, not just documents, according to Kus. He highlights why identity orchestration with Okta is critical as enterprises integrate AI agents into workflows... Read more

explore Keep Exploring

What is Box's focus and approach regarding unstructured data? add

What is retrieval augmented generation and how is it utilized by customers? add

What are the key considerations regarding AI security and data access in the deployment of AI systems? add

What future trends are expected in the integration of AI agents with product platforms? add

bolt Powered by CUBE AI

Ben Kus, Box

search

Rebecca Knight

>> Good afternoon everyone, and welcome back to theCUBE's Live coverage of Okta Oktane here in Las Vegas. I'm your host, Rebecca Knight, alongside Jackie McGuire, my co-host, my analyst, my OG for the week. I love it. We're having a great time.

Jackie McGuire

>> We are. This has been a lot of fun, and we got lunch.

Rebecca Knight

>> It's your nerd's paradise, as you said.

Jackie McGuire

>> It is.

Rebecca Knight

>> This is where you, yeah.

Jackie McGuire

>> This is my happy place, which is weird for a lot of people, but this is my happy place.

Rebecca Knight

>> Well, I think that our next guest, this is also maybe his happy place too. Ben Kus, the CTO of Box. Welcome back to theCUBE. You're a CUBE veteran.

Ben Kus

>> Always happy to be here. Thanks having me.

Rebecca Knight

>> So as you've been on theCube a few times before, but why don't you start to tell our viewers a little bit about Box and what you do?

Ben Kus

>> Sure. So for Box, our whole lives, everything we've done for a long time is all revolves around unstructured data. So in this world, thinking about not structured data and databases, but unstructured data, things like in files and content, images, audio, video documents, these kinds of things that naturally as people we always use and for most enterprises is actually the vast majority of their data, 90% plus. And so our job is to secure it, to collaborate with it, and then of course now to be able to provide AI on top of your unstructured data, which is one of the more exciting things that for us certainly and also for many of our customers.

Rebecca Knight

>> It's exciting, but it's also an enormous challenge that is exceedingly complicated. So talk a little bit about the challenge and the opportunity with AI.

Ben Kus

>> Yeah, so there's this interesting dynamic where in the world of structured data, in the world of machine learning and these various models that you had in the past, there's been a lot of evolution of the past 10 to 20 years, but because it typically would require a person to sit down and understand, you got to watch the video to understand it, you have to read the document, and then there's a lot of nuance built into it, but now with generative AI and AI agents, they actually have the ability like humans to actually understand and to help you create this kind of content, and that used to never really be possible. And so incorporating these AI agents can do things for you, have them do work for you, have them do research, have them structure unstructured data, have them participate in workflows, have them go do multi-steps that you'd say, "I want you to do this, this, this." And they'll just be able to go off and do it and read and understand things. And so this kind of area for most of our customers is something where when they see generative AI, they say, "I have this huge amount of data that I can now consider using and be some of my most critical data and I can get way more out of it."

Jackie McGuire

>> Now, Box is a phenomenal technology. I have known box for a very long time. I was telling you before this, I worked at SVB a hundred years ago and we were Box's first bank, and so I met the team very early on. Box is kind of a blessing and a curse, right? Because this ability to store basically unlimited amounts of data in the cloud is phenomenal, but you don't have to clean it up the way that you do when you run out of storage space on an endpoint. And so it does tend to lead itself to this crazy amount of unstructured data. And I think coming from AI and the data science world, classification is one of our hugest issues and this kind of swamp of, to your point, a hundred different kinds of files. So where does AI come in that whole tugging push and pull between unlimited storage and mess.

Ben Kus

>> Yeah, there's so many places and you're totally right that there is this dynamic where the more stuff you have, the harder is for you to find and to organize and to do anything going forward. At the same time, most customers, they often don't want to get rid of some of their old stuff. It's still valuable and it's still useful in many ways. So for AI, it helps you in a couple of different ways. You mentioned classification. So we've actually just announced our AI classification where the AI can go through and basically be able to classify things and be like from a security perspective or even just to be able to tag data that you say, "I'm interested in these kinds of things." And interestingly, this has been a really hard problem for enterprises because it was kind of hard. You used to take a security professional to look at something and be like, "Yes, this one's confidential according to these rules." But then AI is intelligent enough to actually go through and do that. So we have AI agents that can tag your data to be able to classify it for security purposes. But even more importantly I think is that AI can help you find and process different types of data. And to exactly your point, one of the things that's very useful is that normally if you were going to search for something like you alluded to, you would go search and all enterprise search suffers compared to internet search in that they have all the signals because the things you have access to probably only you have access to them. So then you couldn't get all of the ranking is always slightly harder problem. But you as a person, what you do is you go through, you look at it, you click on it, you see if this is what you wanted, and now you can actually have AI, Agentic AI do that for you, which is basically to perform the searches, then find the right data, not just find the right documents, but find the right data for you in a lot of different documents. And that's becoming one of the new and interesting ways that you would deploy AI to just help a normal person in their normal day-to-day work.

Rebecca Knight

>> That's amazing. So do you have a case study in terms of a customer that has actually said, "Okay, I need that." And then use that insight to make a business decision?

Ben Kus

>> What we see is that this idea, so it's called retrieval augmented generation for the base set of the technology around it. And so what we see from many of our customers, this is one of the, AI could do so many things, and this is the beginning step. And so what they're doing is they're just taking a lot of their data and they're saying, "AI, can you find me this information?" And so we see things like customers who use it for sales, like our sales team at Box is huge fan of this, which is if a customer has a question, you just ask AI and it is able to look through tens of thousands of files, hundreds of gigabytes of data and then come up with not just here's a file to read it, but here's the answer. And then if it finds multiple things with multiple different answers, it's like, "Here's different approaches." Oh, you want a case study? Here's six case studies. The question you ask, I could actually ask AI and it would come back and show me all of those different pieces of information. So you could do this for internal HR reasons, you could do this for sales reasons, you could do this projects and so on. It's just this idea of having AI find information for you is really one of the beginning uses of AI. But it's one of the things that many customers are thinking is like, "This is great."

Rebecca Knight

>> And it's a really great example of how an agent is not just a basic bot, but it's actually collaborating with the human and working together as a thought partner.

Ben Kus

>> Yeah.

Jackie McGuire

>> Trying to understand intention versus just instruction.

Rebecca Knight

>> Exactly. Yeah.

Ben Kus

>> And this is, I think in the world of agents, having agents do multiple steps for you. Previously, there was some of the earlier agents or some of the earlier just models. You ask them something and they kind of do the best they can and they give you an answer. But with the idea of Agentic AI, they go, they'll look stuff up, they'll do research, they'll take them a while, they can look through all your data, they can make a little plan, they can double check it. So it's almost like you hired somebody, a consultant to do all this work. And this is some of the power of AI agents that are coming out now, and I think there's going to be just nonstop improvements in this kind of area.

Jackie McGuire

>> Yeah. Do you see a future? So one of the things that I always lament is we do a terrible job in the enterprise at controlling data creation. So I probably have, and one of the interesting things I've seen coming out of a couple of the AI companies research-wise is that AI doesn't actually understand time very well. So the concept that 2017 information is not as good as 2025 information, do you envision a future where we can start deduplicating data and identifying-

Rebecca Knight

>> Deduplicating data, I like that. Yeah.

Ben Kus

>> You got to dedupe. Yeah.

Jackie McGuire

>> Because when you're trying to use AI, if I accidentally have 63 copies of the same file, that volume, and I think we talk about bias and people think we're talking about demographic bias, we're talking about mathematical bias, and that form can now appear like it's 63% or 63 times more relevant than one file. And so is there a future where we can solve this problem?

Ben Kus

>> I think definitely. So the retrieval step on the retrieval, then augmented generation is something that you definitely want to do that where you go through and you basically are finding the most relevant data, but then also try not to confuse the model by giving it too many other things. This is a very interesting world where it's like a subset of what you're talking about is that finding the most authoritative information is actually another problem. And this is where it's very hard, but as one of the tricks you can do is ask an agent to reflect on it, have an agent go through, find all the data, then think about it for a while, double-check through a chain of thought reasoning so that it's able to come back and be like, "I think because you asked this, I had an old one, a new one, this other one was marked draft. And then so the right answer is probably this, but here's all the reason I believe that, so that you can go double-check that." Then that's sort of the power of these agents can do that now. And that's one of the reasons why people are starting to use it more because instead of it just saying, "Here's something." It says, "Here's something." And a lot of reasoning when I feel like this is a confusing or a non-authoritative topic built in.

Jackie McGuire

>> And it can have a big impact, right? Because if I'm an engineer looking for information and I find let's say an incident report with root cause analysis. At the end of root cause analysis, you usually have next steps. But if those next steps are not actually what ended up getting implemented in the new guidance because of that, you may be following wrong things, right?

Ben Kus

>> Yeah, absolutely. All sorts of great things, including at some point, most of the engineers that I know, they hate filling out those things because it takes a long time. And if you can have AI, you can give it all the relevant data and then it can fill out the kind of official form because it's seen references, it understands the details that you've given it. Those kind of things are just, that's one example of probably hundreds or thousands within an enterprise where they're just AI can help them do their day-to-day job.

Rebecca Knight

>> So we're here at Okta Oktane. Why don't you tell our viewers a little bit about how you at Box works with Okta?

Ben Kus

>> Yes. So one of the key things in all of this that we're talking through is the idea that if your AI isn't secure, you have a serious problem and you just want to deploy it. In our world for unstructured data, one of the big challenges is that everybody has access to different things and therefore you have to be very clear on who when the AI is talking to you, it only can get stuff that you have access to, AI doesn't keep secrets. So if it comes across some data, it will happily tell you. And so you cannot give it access to something unless you have access to yourself. And that idea of identity and that idea of permissions is really critical. And to make it works, you have not just the unstructured data from Box, you have structured data from another system, you have CRM data from another system, you have data from HR system, and oftentimes some of the most useful things is to get many of these different agents working together. In order to do that, you need to work together in a standardized framework. And this is what Okta is one of the key partners on, is to be able to not only provide to you a standard identity, so you know the identity of the people and the identity of the agents, but then also be able to orchestrate it so that you're very clear about which of these agents have access to what. So if they do get compromised or if they maybe get tricked in different ways, then you at least have some controls on what they can do.

Jackie McGuire

>> I think that's really one of the powerful things about at least natural language processing AI is the intent versus instruction. So in some future state, Okta says to Box, "Hey, this user is intended to have access to these types of information." They perform a query in Box, and then an agent says, Hey, three of the documents that came back, you're not supposed to have access to that type of information, even if it's in your Box account.

Ben Kus

>> Yeah, I think this is one of the really interesting areas where previously, if you had a security professional, they could maybe do that kind of stuff for you, but no way anybody could ever afford having that kind of analysis per person, per document done, it's just too much data, but AI agents can do that kind of thing. So it's one of these areas where you can get a much better outcome, reduce risk, improve the efficiency overall, and have it be something that would've never really been practical or possible before.

Rebecca Knight

>> So looking ahead, where do you see the biggest opportunities for Box to make AI easier and smarter and more fun to work with too for developers?

Ben Kus

>> I think we're in the beginning stages of incorporating agents, just the whole industry. So I think you probably, if you looked at the Google trend, you would've seen agents start to appear maybe a year ago. You start to see many companies now releasing their agents, releasing agent frameworks, releasing cross-agent capabilities, and I think we just have a long way to go to make sure that these just get more and more useful. In theory, an AI agent is something that can do really complex tasks, but most people are starting with some of the more straightforward tasks, some of the more automatable things, getting used to it, making sure it performs well, figuring out security. And then from there, they'll just continue to add more and more capabilities for these agents. In the future, it's very plausible that you might be working with more AI agents than you're working with actual knowledge workers. You'll have armies of AI agents to do work for you, and this might be, everybody will become a manager of their AI agents to do all sorts of work for them.

Rebecca Knight

>> A human managing a bunch of robots in addition to actual humans too.

Ben Kus

>> Yeah, and you start to see this in the early phases of code generation where this is how some of the most productive people who use AI code generation work is they're essentially farming out these tasks to multiple agents. And I think you're going to start to see that not just in programming, but also in other parts of the enterprise as these agents get more mature.

Jackie McGuire

>> What can you share from your journey with AI in deploying AI capabilities on some of the most sensitive data that enterprises whole? What can you share with other CTOs, CISOs about that journey? And if they're feeling a little skittish, how can we help them feel better?

Ben Kus

>> Well, so certainly if you're building something yourself, the thing to say is make sure the first thing you've thought of is security, right? Because you just absolutely don't want to be in a spot where your AI is the cause of a serious AI ... security data leakage issues. But then from the perspective of using it, I think what you're seeing is that there's a lot of really good platforms, a lot of really good companies that are coming out with new AI functionality. So like a Box, we specialize, we spend all our time focusing on AI agents from unstructured data, which is this huge world, and then we build it as a platform so that people can build on top of it. And so to me, this is the future that we're going to see is that many different products will be providing you not just their application and their platform, but also AI agents and the ability for you to and build on top of those. So I'm always a fan of if you don't need to build it yourself, then build on top of it. And especially with AI, if you're building it higher up in the stack with AI agents that deploy other AI agents, you're able to take what's the most powerful part of your company and your knowledge to be able to use at that level as opposed to some of the underlying details which you can outsource to your most trusted platforms like Box.

Jackie McGuire

>> I'm a big fan of transferring risk, so it's also a great way to have somebody to sue if something goes wrong.

Rebecca Knight

>> So litigious, yeah.

Ben Kus

>> That's part of the trust is that you trust that companies when they're doing this, they have to think through all of these concerns and therefore, you trust in that they're doing the best that they can out there.

Jackie McGuire

>> Well, I think executives are kind of in this weird murky place where their D&O insurance doesn't necessarily cover AI agents doing crazy things on their behalf, and there hasn't been enough litigation or legislation to say, "Here's where executives are going to be held liable." And so I think you're right that a lot of people are building these things in-house because it's easier to code now, but there's a lot to be said for using the capabilities of a company that actually has a team supporting something, not this gets deployed and then nobody looks at it until the next quarter.

Ben Kus

>> Absolutely. We have a lot of people who focus on all these aspects, data extraction, finding data research, and at some point, we offer it to everybody else and it would seem that they could use those capabilities rather than have to try to put together themselves. In some ways, AI agents are going to be, I think the new applications and that you will see some AI agents are really good and some AI agents are not as good because they have more work to do to make them better.

Rebecca Knight

>> Kind of just like humans.

Jackie McGuire

>> Some of them just there to collect your cellphone information.

Rebecca Knight

>> They don't take PTO, so I don't know, I guess it all evens out. Ben Kus, a pleasure having you on the show. Thank you so much.

Ben Kus

>> Thanks for having me.

Rebecca Knight

>> I'm Rebecca Knight for Jackie McGuire. Stay tuned for more of theCUBE's live coverage of Okta Oktane. You're watching theCUBE, the leader in enterprise tech news and analysis.