Unlocking the Value of Content in the AI-first Era

05:01:13.130 –> 05:01:27.750
Julia Nimchinski: Amazing thanks again, and we are transitioning to our fireside chat with Ben Guts, the CTO. Box, and Jonathan Crawford’s head of growth, and Gtm. At momentumio. Welcome to the show! How are you doing.

1637
05:01:28.910 –> 05:01:30.800
Ben Kus (CTO, Box): I’m doing great thanks for having me on.

1638
05:01:31.000 –> 05:01:32.820
Jonathan Kvarfordt: How long you say, Ben? 1st I’m doing good.

1639
05:01:34.690 –> 05:01:37.150
Julia Nimchinski: Amazing, Jonathan, take it away.

1640
05:01:37.150 –> 05:01:37.710
Jonathan Kvarfordt: Hmm.

1641
05:01:37.940 –> 05:01:53.849
Jonathan Kvarfordt: Oh, my gosh, okay, well, I have the lovely pleasure of being able. I don’t know if I should introduce you, Ben, but I’ve just been doing some research on you, and I think I found 9 patents in the last 10 years something like that. I’ve seen a bunch. You’ve released. One of them, I find fascinating actually is just one more recently around how you released a patent on

1642
05:01:53.850 –> 05:02:15.679
Jonathan Kvarfordt: methods for identifying selected portion of content objects for use in generating Llm. Prompts, which is fascinating. So anyways, I’m like so intrigued and really excited to talk to you. You are the CTO of box, and just been doing a lot of reading, and what you you do and what you’re passionate about. But before I go on and probably say something that’s not accurate. Tell us more about you and your little bit about your journey.

1643
05:02:16.200 –> 05:02:40.159
Ben Kus (CTO, Box): Yeah. So my journey is in the world of, I started. I consider myself technologist. And so I went to Berkeley computer science. I was in a small company that grew up to be a bigger company. Big fix back in the day in Enterprise Security world got acquired by Ibm. Started my own company with some of my people. I worked for a long time, and then that got acquired by box, and I’ve been at box for 10 years now.

1644
05:02:40.495 –> 05:02:47.529
Ben Kus (CTO, Box): and I started the product team. And now I’m CTO, and amongst other things, I want the. My primary focus is

1645
05:02:47.530 –> 05:02:58.130
Ben Kus (CTO, Box): all day and night. AI box. Of course, we’re on structured data. So we focus on the idea of larger enterprises how to handle unstructured data and bring an AI to your to your content.

1646
05:02:58.500 –> 05:03:11.690
Jonathan Kvarfordt: I love it so for those people who have not used box. Can you tell me? And I and some of these questions? Obviously, you and I have gone over, and some I’m just going to ask you, because I’m curious. What are some of the main use cases? You think that box delivers the most value for people, just so they can conceptualize what you do a little bit better.

1647
05:03:12.040 –> 05:03:38.189
Ben Kus (CTO, Box): Yeah. So I’d say, at a high level, like many of us in the industry, have been going through the last, let’s say 1020 years have really seen data become a really critical aspect of almost every company. And so and this is typical of structured data. And so we have a number of technologies, a number of great things, including machine learning, that helps people use data more effectively to the point. Now, it’s like woven in the fabric and the technologies and the great companies that provide this in the world.

1648
05:03:38.190 –> 05:03:55.949
Ben Kus (CTO, Box): But then, unstructured data, like content, like box, has historically always been there. We help you sync your data, we help you collaborate, we help you view, secure, govern, add compliance for your data. But up until the advent of generative AI, we weren’t able to actually fully automate and fully utilize unstructured data in the enterprise.

1649
05:03:55.950 –> 05:04:17.289
Ben Kus (CTO, Box): And it’s kind of compared to structured data. It’s like 90% unstructured is 90% in volume of what people do. And so we have over 120,000 enterprise customers. And typically when we’re talking about AI, one of the 1st things that they’ll think about is how to start to automate some of the things that used to be unautomatable like not possible to do, because there’s no technology that could.

1650
05:04:17.290 –> 05:04:17.620
Jonathan Kvarfordt: Right.

1651
05:04:17.620 –> 05:04:28.660
Ben Kus (CTO, Box): Like, understand this video, understand this document, understand the nuances of the risk of this contract and everything else. And so this is the kind of things for us that people are interested in from our customer set.

1652
05:04:29.920 –> 05:04:51.370
Jonathan Kvarfordt: So do you feel like? And again, we could I go on so many paths with you which I’d love to? But with that in mind. What do you feel like when people come to you or companies come to you with an enterprise? What do you feel like? The common challenges they’re going through saying like, is it? The lack of ability to integrate different data sources is just security, like, what is it you guys help help them identify, and then be able to execute on.

1653
05:04:51.620 –> 05:05:08.810
Ben Kus (CTO, Box): Yeah. So in general, for a lot of our customers who are like again, bigger, bigger organizations, who are in the midst of trying to understand how AI can change the way that they work, because that’s kind of some of the promise of what AI can do. Many times they’re trying to sort of balance this idea of

1654
05:05:08.810 –> 05:05:32.540
Ben Kus (CTO, Box): getting started in sort of practical ways that have value with the sort of the pressure overall to like, try to completely revolutionize everything that they’re doing. And so from our perspective, one of the things to do from AI, but also in general, is to make sure that you’re focused on a couple of your platforms that then give you these capabilities to be able to do AI effectively across your enterprise.

1655
05:05:32.540 –> 05:05:41.319
Ben Kus (CTO, Box): So we kind of see that there’s like a number of data sources that many, many companies have, including things like your structured data, your unstructured data, your Crm, your Hr, and so on.

1656
05:05:41.320 –> 05:06:07.229
Ben Kus (CTO, Box): and then, not only having those be able to have AI themselves, and to do things like Agentic AI, and some of the newer promises of the way that you can automate more and more with this kind of newer generation technology, but then also be able to coordinate those with other tools out there. There’s a number of sort of like layers of agents that, like at a higher level, either enterprise, build themselves or other 3rd party tools that can kind of like coordinate across all those. So most of the customers we talk to. They’re in the middle of this journey of

1657
05:06:07.270 –> 05:06:18.150
Ben Kus (CTO, Box): solidifying the basics, making sure it’s secure, making sure that it’s it’s compliant, making sure they can do specific things, increase the productivity of their users, and then also then try to automate some of the higher level enterprise processes.

1658
05:06:19.240 –> 05:06:43.669
Jonathan Kvarfordt: One of the things you brought in, if I remember correctly, is Box AI studio right? And part of it was to make sure you could help create agents across different workflows and use different models like there’s a lot of cool things you do with the studio. I’d love to have you tell me more of your thought. Obviously, as a technologist expert, I feel like people talk about agents a lot. But what’s being missed a lot is the importance of the data that agents need to really get

1659
05:06:43.670 –> 05:06:56.190
Jonathan Kvarfordt: good use out of AI power. So tell me more about like how box studio, or just your thinking in general, how do you identify what data is needed? So you can really release agents or a fleet of agents, whatever you want to call it in the best way.

1660
05:06:56.530 –> 05:07:18.220
Ben Kus (CTO, Box): Yeah. So in general, when when we’re talking about agents, we’re often talking about the the sort of the concept that AI can not only do something for you, but can do more and more complex tasks that have, like. You know, objective instructions. An AI model to power it tools to go access different things. So from that concept of an agent.

1661
05:07:18.220 –> 05:07:30.800
Ben Kus (CTO, Box): one of the the more critical things is that it can be able to access your data across the board and been doing it in a secure way, like, if you think about it, like to the extent that agents can resemble sort of like the power of humans.

1662
05:07:30.800 –> 05:07:55.310
Ben Kus (CTO, Box): Imagine that you get like the most intelligent employee in the world. They got 20 Phds. And they’re very willing to work really hard, and they show up in your company. And then, if you don’t give them access to what you have internally like, you know your tools, your systems what they work. They’re just not be as useful as they could have been otherwise. And so then, naturally, then, one of the goals is to hook up the AI agents to your data.

1663
05:07:55.310 –> 05:08:23.500
Ben Kus (CTO, Box): But then, if you do that in sort of if you’re not careful, like, you have a giant problem, because, like, you know, that’s like. For most of your enterprise systems like not everybody has access to everything out there, right? Like the stuff that any one person, any enterprise. Access to is just different, you know, not only because you’re on different teams, and you have different roles, but also you have your stuff. You have stuff. That was a specific project you have access to, maybe something that’s confidential about your you know your manager employees, or whatever else. And so you have to be able to hook

1664
05:08:23.830 –> 05:08:32.669
Ben Kus (CTO, Box): the way these things work. And then, in particular, the access and of the permissions and the user base access controls.

1665
05:08:32.670 –> 05:08:56.660
Ben Kus (CTO, Box): And so those are the kind of things that you need to not only just have AI do stuff on, but also to make sure that your AI is not your number one challenge of data leakage and security issues. You need to actually securely do AI on top of your data. And that’s often one of the key things that either prevents customers from like starting the adoption, or actually is something that is like their biggest problem solved. To try to even begin to use some of this stuff

1666
05:08:56.660 –> 05:09:08.560
Ben Kus (CTO, Box): so that box one of our let people do secure AI on their content across not only the models that are trustworthy, but then also the ability to map directly to the user access controls.

1667
05:09:09.110 –> 05:09:15.240
Jonathan Kvarfordt: So you feel like you’re more of like an orchestrator kind of moving things. Or is it more of just an integrator, of making sure it connects different different dots.

1668
05:09:15.983 –> 05:09:21.093
Ben Kus (CTO, Box): A a little bit of both like the the idea of like

1669
05:09:22.100 –> 05:09:36.880
Ben Kus (CTO, Box): 1 1 of the interesting things is is when you’re when you’re like thinking about AI like, and then having access to like content like you start to to see that the AI starts to have some of the challenges that like what people would have about like like, how do I give it access to the things I want? How can it like.

1670
05:09:36.880 –> 05:09:37.450
Jonathan Kvarfordt: Go ahead!

1671
05:09:37.450 –> 05:09:55.020
Ben Kus (CTO, Box): In the right format like, how can I? How can I smoothly make sure that, like new stuff it has access to or not like until you start to go down this path of like, well, I need to have retrieval augment generation or secure retrieval augmented generation. I need to have the vector databases and store the embeddings and be able to look up info across a big corpus of data. You know, some of our customers will have like

1672
05:09:55.020 –> 05:10:19.809
Ben Kus (CTO, Box): dozens or hundreds of petabytes of data like in that we we help and like, that’s just not a trivial thing to be like. Oh, now have AI, go do something with that, like, you know, not like for security reasons, but just practical infrastructure reasons. And so this is an area where like what we focus on is to bring AI to your content in a way that you don’t have to worry about all stuff. If you want the benefits of a unstructured content, which we think is one of the biggest benefits of in the enterprise. Then you need a mechanism, a platform to be able to handle that.

05:10:20.900 –> 05:10:41.650
Ben Kus I feel like a lot of times when it comes to the subject. People don’t know what they don’t know, and don’t even know what’s possible, even if they did know it, you know. So it’s like, how do you help companies identify what kind of unstructured data should be the priority to salt, start structuring, and then to give more ability to the leaders and teams actually execute like, how do you identify where to even begin.

1674
05:10:41.900 –> 05:11:01.989
Ben Kus (CTO, Box): Yeah, so 1, 1 of the things that lives as interesting when you start to look into not just the the personal productivity types, use cases which are important, like being able to search, be able to get answers to questions, being able to to like, have AI kind of assist you when you’re viewing files and things like, but but into the world and more business process, like one of the the key questions, is always like, What’s the slow part

1675
05:11:01.990 –> 05:11:12.362
Ben Kus (CTO, Box): of whatever you do internally, and many companies have like gone through. And they’ve automated an awful lot of things like so we were dealing with one of our customers. We were helping them out on this exact kind of thing, and they and and and they were

1676
05:11:12.570 –> 05:11:14.850
Ben Kus (CTO, Box): pointing out that they have a loan origination process.

1677
05:11:14.850 –> 05:11:29.230
Ben Kus (CTO, Box): And the slowest part of the whole thing was to actually review the data that the person had to upload to prove that they’re like who they were, so you know, like utility bills and personal documentation, and so on. And they had to have, you know, so they’d uploaded. It had to have somebody review it, look at it like double check 10 things.

1678
05:11:29.230 –> 05:11:50.619
Ben Kus (CTO, Box): And so then, and the reason why that was the longest is because it was the thing that wasn’t automatable. But this actually like, would lead to like in loan processing. And so this is an example of many different examples across industries where you could have AI in this case, in the form of an AI agent, go like, actually look through and be like. I know my objective is to, you know, understand this document, to follow these these rules.

1679
05:11:50.620 –> 05:12:13.710
Ben Kus (CTO, Box): and then to identify to the the person who’s responsible, approving the agent or the approving. The loan is, is like, what is is this in line? Does it have the right name? Does it have the right dates? Does it have the right, everything. Look at an order, and then those kind of things help automate. And then also, just like not not just save people. Time like internal employees, but also just speed the process of something that was otherwise taking a lot of time. And we see a lot of examples like this across.

1680
05:12:13.710 –> 05:12:39.850
Ben Kus (CTO, Box): Like, now, we have a travel agency. We have a lot of financial examples, people processing these things that in some cases still paper, they get scanned. Yeah, like, not everybody in the world has fully gone to like structured data systems. And so, interestingly, AI gives you a chance to almost leapfrog and be like, look! Don’t spend all your time investing in a bespoke custom application. Do it? Just have AI. Take a look at it, and it’ll start to solve some problems for you. And then this is the kind of thing that many of our customers look into.

1681
05:12:40.820 –> 05:12:56.360
Jonathan Kvarfordt: You feel like there’s common again, partly because people don’t know they’re not AI experts, not automation experts. There’s probably a lot of missteps or things they think they can do so across, whether it’s enterprise or mid market. What are like common things you see, people either focus on, that’s

1682
05:12:56.590 –> 05:13:05.300
Jonathan Kvarfordt: I don’t know. I don’t wanna say wrong. I’ll say not the most effective. What? What’s the missteps that can avoid that you’ve guys learned through, you know, obviously working with thousands of customers.

1683
05:13:05.570 –> 05:13:34.530
Ben Kus (CTO, Box): Yeah, so there’s sort of a there’s like a couple like you know, every every single company we talked to is quite aware of the idea of AI can do an awful lot for you right? And then some of them, I think might be like like there’s this, this concept of like, how do you start and like, how do you make this work? So let’s say you have a process that is like, you know. Multi step process. Maybe like that loan origination or some other process like at some point, somebody says, in order to make this truly recognize the value. I’m gonna replace the whole thing.

1684
05:13:34.660 –> 05:13:47.279
Ben Kus (CTO, Box): AI seems like, do everything. So I’m going to make it do everything. But then, when you kind of decompose it, you’re like, well, that that thing which was like, you know, maybe 10 steps each step itself was a very complex, very critical thing to get right.

1685
05:13:47.280 –> 05:14:12.240
Ben Kus (CTO, Box): And so rather than trying to automate everything across the board. And the 1st shot, typically what we recommend and the things I’ve seen be more successful is like, kind of like, go down to the bottom of your sort of agentic or your internal process, and then see like this part here is, if I can save this, then I, and maybe it’s the most repeatable or the most automatable piece of it. Then I can then build many of these and eventually start to

1686
05:14:12.240 –> 05:14:22.790
Ben Kus (CTO, Box): them overall. And actually, what we’ve seen is it was interesting is that many of the things that people start on when they take this approach are the least popular, slowest, most sort of onerous part of a process that nobody wants.

1687
05:14:22.790 –> 05:14:23.680
Jonathan Kvarfordt: Right.

1688
05:14:23.680 –> 05:14:33.310
Ben Kus (CTO, Box): And then and then being able to have AI agents do that. You know, it’s actually interesting, because sometimes they can just do a better job simply because they you know. Sometimes people don’t like that part.

1689
05:14:34.240 –> 05:14:58.589
Jonathan Kvarfordt: Yeah, do you feel when you see people rolling out strategies? And I have. I have my own thoughts on this question. But I’m curious. Yours, you see, like the shopify CEO saying, Hey, everyone needs to use AI. You gotta justify AI. So that’s kind of more of a bottom up force approach, and then the other side is kind of more across enterprise motion. That’s top down, you know, because you need more of a strategic multi team. Look at something which, like box sells for

1690
05:14:58.590 –> 05:15:09.190
Jonathan Kvarfordt: where? Where do you feel like the balance is between? Because I think it’s good to have both. How do you look at? When when do I empower the individual versus? When do I need to look at more of a cross organizational, larger.

1691
05:15:10.240 –> 05:15:31.689
Ben Kus (CTO, Box): I mean, there’s an interesting dynamic, which is that like, typically, if you ask, like a group of people, and to say, like, how many of you have used generative AI like Chat Gpt or Gemini, or any of them, like everybody, raise their hand right like. There’s not a lot of people like. I’ve never been to use this before. And then but if you ask them how many people have used it in the enterprise like you get like scattered shot. And then, if you ask why? It’s like, well.

1692
05:15:31.690 –> 05:15:47.029
Ben Kus (CTO, Box): it’s not approved for security. Somebody fears something we haven’t identified for sure. The Roi on it or whatever. And so I think one of the key things that we’ve seen be successful is that when the leadership focuses, not on trying to dictate the use case that matters the most.

1693
05:15:47.050 –> 05:16:02.350
Ben Kus (CTO, Box): but rather on removing some of the barriers that then allow the people to use it. The way that they use it in their internal lives. And so, of course, you need security. Of course, you need to make sure that your data is protected. Of course you need to make sure that you have the right systems in place, and you’re not doing irresponsible things.

1694
05:16:02.350 –> 05:16:10.599
Ben Kus (CTO, Box): But then, once you get that out of the way, and you approve. And you say these are the kinds of platforms and the kind of capabilities that will allow. Then what we’ve seen is that many times

1695
05:16:10.600 –> 05:16:33.130
Ben Kus (CTO, Box): the thing that maybe the higher level leaders thought were the most interesting or solve the biggest problems are actually different than what the people have found the uses. For because who knows better how to help the workers be more productive, and to do more for the customers than than those people? And so then, so to me, the number one leadership role would be to remove the barriers that prevent people from doing AI

1696
05:16:33.130 –> 05:16:49.089
Ben Kus (CTO, Box): and then to then encourage experimentation. Sometimes, also, like, essentially manage big projects, especially across teams. But but often, just like you get more out of letting people use it than you do from like trying to guess is, what’s the best use.

1697
05:16:49.950 –> 05:17:07.119
Jonathan Kvarfordt: When you, because security is usually one of the reasons why I hear people have a challenge with this, at least organizationally, because they don’t roll out Gemini, whatever the tool might be, because of security concerns, I know there doesn’t exist yet. I know it’s on the table, I think on I was reading an article the day of doing some sort of AI specific data security

1698
05:17:07.230 –> 05:17:20.589
Jonathan Kvarfordt: measure. But obviously right now you have Soc 2, type 2, and you have Iso 2 0 2, like, what are the things you suggest to people to say? If you’re looking to do a tech that’s cross-functional, what are like this table stakes. You have to have this that you would feel comfortable with as a CTO.

1699
05:17:20.920 –> 05:17:46.260
Ben Kus (CTO, Box): Yeah. So I mean, one of the interestingly, the 1st questions about giving your data to AI is is actually the same question you’ve always had of giving data to anybody. Which is, do I trust that company to have my data? Do I have a contract with them? What does this say that people can do one of the scariest things even again, not even not even really. AI is when you’re giving your data to something that you’re not sure how they make money. You don’t have a contract with. You’re not sure their terms of service.

1700
05:17:46.260 –> 05:17:46.810
Jonathan Kvarfordt: Right.

1701
05:17:46.810 –> 05:18:06.980
Ben Kus (CTO, Box): Then oftentimes what that means is that like part of the reason they’re giving you a free service, or that they’re they’re they’re like providing this is that they’re going to do things that maybe you’re not comfortable with with your data. And in the world of AI this is always a concern. This is what makes kind of people like extra cautious is that like some AI systems will go. I mean, they’re they’re open about it. It’s like a little checkbox that says, if you don’t

1702
05:18:06.980 –> 05:18:16.920
Ben Kus (CTO, Box): check this box like we’re going to like, learn about your data right? And that’s like one of the scariest things in the world is to think I have proprietary data. When my employees, you know, ask a question online, they click thumbs up.

1703
05:18:16.920 –> 05:18:17.480
Jonathan Kvarfordt: Right.

1704
05:18:17.480 –> 05:18:30.069
Ben Kus (CTO, Box): Answer, you know, like what is my employee? Info? And what is the risks associated with this financial thing, or whatever? This is a great answer. And then suddenly, that turns into data that goes in a train run. And now a future model from whatever vendor actually has that data.

1705
05:18:30.200 –> 05:18:52.849
Ben Kus (CTO, Box): Now. So then, that’s very plausible to do. Now, most enterprise providers of AI will like strictly say no like that’s absolutely not going to happen. So like a box. We we will explain great detail, both contractually and also technically like, why, that doesn’t happen. We we have no interest in in training on your data or let anybody else do it. Our job is to facilitate the use of AI on your data.

1706
05:18:52.850 –> 05:19:01.809
Ben Kus (CTO, Box): And so this is the kind of thought process that I think most people need to go through is to say, What do I need to do with my like? How can I trust this company?

1707
05:19:01.810 –> 05:19:24.460
Ben Kus (CTO, Box): And then, of course, a box. This is just what we do. We store your data, we let we upload download it. We let you collaborate on it. And all these things. And so it’s not a long leap of logic to say, and I can do AI on it, and we’ll tell you about how we do? AI securely, but in general make sure you trust your vendors and make sure that they have AI terms and principles you agree with. Make sure that you’re sure about how they log and maintain your data

1708
05:19:24.460 –> 05:19:33.480
Ben Kus (CTO, Box): and and make sure that you have control to be able to turn off and on things in some cases, if you have preferences for models, switch those and so on. So those are the kind of the the major areas. AI trust.

1709
05:19:34.480 –> 05:19:48.539
Jonathan Kvarfordt: Yeah, I love that. I I. This last year I learned how Openai offers very limitedly a 0 attention data policy, that kind of stuff, and not every vendor has that. So it’s good to know, like, who has what agreements with which Llm. Model to know how the data is being used. You know.

1710
05:19:48.540 –> 05:19:49.159
Ben Kus (CTO, Box): That’s great for us.

1711
05:19:49.160 –> 05:19:49.480
Jonathan Kvarfordt: Perfect.

1712
05:19:49.480 –> 05:20:13.490
Ben Kus (CTO, Box): Like again, we can’t. We can’t. Like people have data. Residency concerns. People have governance concerns, people have encryption concerns, and and all those kind of got the window when data is floating in different places. And and so this is why we have the. It’s important to us is to not let anybody, including our some very trustworthy companies out there, but just still, like contractually, not let them say that they’re gonna retain that data because it’s the customer’s data, and we need to, you know. Make sure we’re good stewards of it.

1713
05:20:13.660 –> 05:20:28.379
Jonathan Kvarfordt: Yeah, in the last 8 min. Because I have a feeling, you can go on this topic for a while. You are on a podcast just recently, we talked a lot about systems thinking and specifically like it is individuals and also as organizations. And I don’t think people.

1714
05:20:28.560 –> 05:20:45.449
Jonathan Kvarfordt: I went through a systems thinking class in an Mba. Which was fascinating. But it’s not a topic that I’m a i could say an expert in. I’m learning more about it. So because this is something you’re passionate about like for someone who’s either building out their own agents, their own workflow, or an organization doing it more across orgs.

1715
05:20:45.610 –> 05:20:53.850
Jonathan Kvarfordt: What are the principles around system? Thinking that, you say are like critical to make sure you can really fully leverage agents, the fullest capacity.

1716
05:20:54.360 –> 05:21:05.410
Ben Kus (CTO, Box): Yeah. So I’ll give you a slightly different answer, for if you’re building it yourself versus if you’re an enterprise adopting platform. And so for for the to the point of when they’re like building things themselves

1717
05:21:05.690 –> 05:21:26.340
Ben Kus (CTO, Box): like. If I went back in time to myself like 3 years ago I would say, Hey, make sure that the very 1st version you start to build agentically like, and by that I mean that not just consider AI to be like a 1 shot thing like an AI model that, like, you know, returns something. But to have it. The AI grow in how it can orchestrate the workflow.

1718
05:21:26.340 –> 05:21:55.930
Ben Kus (CTO, Box): And so like, there’s a lot of great agentic frameworks out there. And I think any of them are good. But basically allowing your development process and allowing that the way you think about things to be like, oh, this! This is not as accurate as I wanted. It’s not doing thing I want. I know I’ll have AI 2 steps instead of one or 3 steps or 4 steps, and then to be able to switch out models. Take advantage of these capabilities like reasoning models and models that put out structured content and models that could do multimodality. Those were things that weren’t obviously clear to 3 years ago.

1719
05:21:55.930 –> 05:21:56.610
Jonathan Kvarfordt: Right.

1720
05:21:57.000 –> 05:22:25.419
Ben Kus (CTO, Box): And so then what we had to do along the way was retrofit, our whole system to not just have a bunch of functions and a bunch of processes together that, like, you know, tried to get that data in a format to give to an model, but rather have a true agentic system that could traverse workflows and doing that actually helps us a ton to be able to do things like just adapt to the latest capabilities. Because there’s a funny thing that happens where like, it seems like now, almost every 2 weeks. What is the best model in the world just becomes something else? I don’t know Gemini, or

1721
05:22:25.420 –> 05:22:48.800
Ben Kus (CTO, Box): it’s the new open, a model. It’s the new anthropic model. But like in in at some point they became able to do something that wasn’t really possible like, not even that long ago, 6 months. And then so you need to have the ability to then say quickly, Well, now, that’s possible. I’m going to then be able to provide that use case and not have it. Be a bunch of development work to be able to turn on releases. See? This idea of platform agentic thinking is something that I would encourage.

1722
05:22:49.150 –> 05:23:04.080
Ben Kus (CTO, Box): not just Isvs to do when they’re thinking about like an AI 1st or an AI native capability. But then also companies to when they’re thinking about their their data and their platforms, and how they do things like make sure that all everybody you work with is thinking.

1723
05:23:04.400 –> 05:23:16.379
Ben Kus (CTO, Box): You know. They say the the you know they talk AI first, st they they think AI first, st they’re able to provide you functionality as AI, first, st because everything’s evolving so fast that you’re almost just rather than betting on any one

1724
05:23:16.490 –> 05:23:25.510
Ben Kus (CTO, Box): like AI technology or any one. AI use case. You’re almost betting on people and platforms that are like able to express that for you according to your enterprise. And needs.

1725
05:23:26.370 –> 05:23:45.129
Jonathan Kvarfordt: Awesome. So for the individual. Again, this whole topic around system thinking, not just from a technology standpoint, but just like there’s a difference between processing and then systems. Right? So what would be like some takeaways? What would be the top? I don’t know. 1, 3, or 5. You pick takeaways around, how do I build system thinking? So? I can really leverage agent workflows.

1726
05:23:47.360 –> 05:24:03.259
Ben Kus (CTO, Box): it’s a good question, I I think. The some of this would be to enable platforms that can do capabilities and then build use cases on top of those build in security and compliance from the beginning. So you don’t have to go retrofitted afterwards.

1727
05:24:03.260 –> 05:24:23.700
Ben Kus (CTO, Box): And also, you know, to continue to experiment on enabling people across the board to, to utilize, to, to quickly use AI in different ways like. So these are the kind of things that I think are the most important, as you consider, like different aspects of how your your company, or how your enterprise will be able to.

1728
05:24:23.700 –> 05:24:24.060
Jonathan Kvarfordt: Right.

1729
05:24:24.060 –> 05:24:25.529
Ben Kus (CTO, Box): Utilize AI in the future.

1730
05:24:26.600 –> 05:24:39.939
Jonathan Kvarfordt: I’m curious. I’m now gonna ask you a personal question. If you don’t mind me asking what is the either most advanced agent that you personally use, whether it’s personally or professionally. And then what does your team use? That you would be like? This is so awesome. What is that thing.

1731
05:24:40.490 –> 05:25:06.950
Ben Kus (CTO, Box): I will give you an answer, which is, that I continue to almost every day be impressed as to how many things you can do that you can represent as a little agentic workflow. And so this is like a revelation that, like these little State diagrams that you like sometimes see in these academic papers, or that you maybe learn in school, or whatever it’s just like, do this and this, and a branch here, or whatever like, because everybody’s aware that, like AI can do this one thing, and then maybe they’re going to where it could.

1732
05:25:06.950 –> 05:25:07.380
Jonathan Kvarfordt: Right.

1733
05:25:07.380 –> 05:25:14.730
Ben Kus (CTO, Box): Steps. But then, if you have AI coordinate of those 3 steps like it’s, it’s weird, is like there’s suddenly stops to be like

1734
05:25:14.760 –> 05:25:36.728
Ben Kus (CTO, Box): and introduce tools so they can go get and change things. And suddenly, you’re like, Huh! This can do an awful lot of things like to the point of of like, you know, starting to you start to see the promise of AI. How it can work in the enterprise is by constructing these little workflows. Now today they’re not the easiest thing to build you can. They can go wrong, they can get distracted and so on. So so this is why you you spend time and effort to like

1735
05:25:36.970 –> 05:25:59.769
Ben Kus (CTO, Box): get them to focus and to actually solve the problem at hand. But that, like across multiple vendors, across multiple technologies across everything is like rapidly getting better. So I kind of foresee a world where the thing that’s like stopping you as an enterprise or as a person from getting something done with AI is your ability to explain precisely what you’re looking for, and then that’s something I think people kind of pick up over time, or they have nice templates that come from companies

1736
05:25:59.770 –> 05:26:07.959
Ben Kus (CTO, Box): like box or others. And so I see that as sort of the the thing that continues to fascinate me, and and takes my focus like almost every day.

1737
05:26:09.468 –> 05:26:26.110
Jonathan Kvarfordt: You know. What’s funny is that as I teach AI to different people at university level and just my kids, I’m always fascinated by how AI forces you to be a better communicator like you said with context or just instructions. It makes me be very self-reflective of like, how am I communicating to get what I need to get, you know.

1738
05:26:26.331 –> 05:26:42.480
Jonathan Kvarfordt: One last thing is because you’re a very unique spot, kind of talking about the future of AI, and what you think you should prepare for. You talked about how, if you go back 3 years like, what do you project your 3 year person ahead of you is going to come back and tell you I was like, this is what you need to prep for. What do you think it’s coming.

1739
05:26:43.771 –> 05:26:58.860
Ben Kus (CTO, Box): My my best guess would be that for the kind of our customer set that we have is that we not only need to provide to them the basic building blocks so that they could then orchestrate whatever they want themselves on their content, but then also provide to them

1740
05:26:58.860 –> 05:27:21.360
Ben Kus (CTO, Box): predefined templates that would help them get sort of most of the way there like, and so many of the like more advanced use cases that our customers are asking for today, we’re like we could totally build that. We have state diagrams and workflows and all these great things, and a secure way to access your data, but and then and then we can. And we have, like, you know, Sis, or consultants, or whatever. But at some point it maybe is for us to like, give them something that’s mostly there, so they can customize it to go forward. And so you get to the point where, like, you basically make

1741
05:27:21.360 –> 05:27:27.464
Ben Kus (CTO, Box): enterprise it. People able to accomplish a lot more without having to go too far down the stack to to

1742
05:27:27.810 –> 05:27:29.500
Ben Kus (CTO, Box): figure out how everything works.

1743
05:27:30.140 –> 05:27:45.200
Jonathan Kvarfordt: Yeah. Do you see where I’m gonna ask one more question. Sorry, Julia, do you see a world where Openai talked about the 5 levels of where we’re gonna go. And one is the organizational agents. Type thing, I mean, I see box being a platform that could like essentially enable that, you know. You see that kind of world coming.

1744
05:27:45.700 –> 05:28:08.929
Ben Kus (CTO, Box): Yeah, I think like, there’s definitely an interesting world as you get, not only like these platforms and these these agent capabilities, but as they roll up into like almost any like individual workflow that’s complicated can be rolled up into a bigger agent. So like at some point, you might have these, like very high level agents that are able to do an awful lot in drawing upon different systems, like box or different other agents, and like, I think that is definitely something that we would be seeing.

1745
05:28:09.900 –> 05:28:14.165
Jonathan Kvarfordt: That’s fascinating. Thank you, Ben. Julia’s giving me the the cane. I gotta stop talking.

1746
05:28:14.450 –> 05:28:18.439
Julia Nimchinski: Question. Thank you so much. Again, Jonathan Ben.