AI Integration: Box is leveraging AI agents to automate workflows and extract insights from unstructured data, significantly enhancing productivity across various job functions.
Product Launches: Recent launches include Box Extract, Box Automate, and Shield Pro, which facilitate the use of AI agents to process and manage enterprise data efficiently.
Market Position: Box is strategically positioned to capitalize on the growing demand for AI-driven data management solutions, benefiting from its established trust and infrastructure in handling unstructured data.
Industry Impact: The integration of AI into enterprise workflows is expected to transform business operations, with Box enabling companies to unlock the value of their data through advanced AI capabilities.
Competitive Advantage: Box's ability to offer customizable AI agents and its model-agnostic approach, using technologies from OpenAI, Anthropic, and Google, provides a competitive edge in the AI application space.
Financial Growth: Box's recent revenue growth is driven by increased demand for its advanced AI features, with a focus on expanding its Enterprise Advanced plan that includes AI-driven tools.
Future Outlook: The company aims to further enhance its AI capabilities, addressing current limitations like context window constraints, to fully automate complex workflows and expand its market presence.
Strategic Focus: Box is committed to continuous innovation and execution at a startup pace, aiming to maintain its leadership in the AI-driven data management sector.
Transcript
So one of the biggest use cases that now we're seeing is customers saying okay I have a million contracts a million invoices and I want agents to go and automate the workflows around that information. So we are executing at the rate of a startup right now at a faster pace than I've ever seen as a company and we are you know we're grinding 24/7. We live and breathe and and sleep AI. Um it's the most fun we've ever had as a company and and I think it's the most important moment for unstructured data and what now companies can do with it in the history of computing. Hello everyone, welcome to another session of Opto sessions. Aaron Levy is back. Since the last chat, Box Works just dropped a new wave of launches. Box Extract, Box Automate, and Shield Pro. And Box's AI agents are now working across Microsoft Enthropic, OpenAI to turn unstructured content into actual outcomes. Aaron, welcome back. >> Hey, thanks. Uh, good to uh good to be on again. >> And are you calling calling from HQ today? Uh yes. Yeah. Here in uh lovely California. >> Nice. Um so yeah, eight months ago you said agents would move from demos to dollars. Uh coming out of Box Works, what's the best customer outcome you've seen from your AI agents? >> Yeah. So for us um we're you know we live in a world where um enterprises have just a tremendous amount of of corporate data in the form of contracts and marketing assets and financial documents and invoices and all of that data contains incredibly rich business information in them. Um but unless a person is sort of on a computer looking at a document you're never really like pulling out the real insights from that data. And so what we've built is a set of tools um that let you take AI agents and have those agents process that information and data at scale. So read through a document and extract kind of critical data points from from that document. So in a contract it might be financial information. In an invoice, it might be, you know, who has to, you know, pay you, you know, the the uh the the um uh you know, the the payment and what's the amount owed. In a um research document, it might be all of the key terms uh that relate to that field of research. And so then later you can query it and synthesize that data. So all of our content has this rich information in it and we've never been able to tap into it. And so AI agents finally let us unlock that. So one of the biggest use cases that now we're seeing is customers saying okay I have a million contracts a million invoices a million research you know documents and I want agents to go and automate the workflows around that information. So this has been a uh really the first year where that became available within our platform. And as you mentioned at Box Works, our customer conference, we announced a new capability called Box Extract that makes it very easy to use extraction agents at scale across your data to pull out all of the most relevant insights from that information. >> Okay. And who so who are the users do you think are going to get the most value out of this? Are we talking about like you mentioned like lawyers that are looking through contracts? >> So So I would say kind of quite literally every every job function in an enterprise, you know, deals with with unstructured corporate data. Um, you know, some more than others, but but every single job you're you're looking at some form of a document or a marketing asset or a PowerPoint or a PDF every single day of work. And so agents that can now bring automation to that type of work mean that every single worker will be in some way aided and and enabled uh to have better productivity because of agents and because of uh agents on their data. >> And so this intelligent content management obviously is a big part of your strategy going forward. Um what's the most misunderstood part of this strategy would you say? Um, I think the, you know, fortunately there's actually a growing understanding of of the strategy and what we're up to. We are incredibly excited about, you know, how customers are recognizing the real value of of this information. So, we have more and more companies every single day coming to us saying, you know, I have all of this data that we generate and I I you know, if you think about any average enterprise, about 90% of their information, 90% of their data is unstructured. So think about structured and unstructured as structured data goes into a database. It's a it goes into a CRM system or an HR system. Unstructured data is contracts and documents and marketing assets. It's all it's all sort of messy mess messy, you know, kind of more human readable data. And computers in the past have never been able to really process the 90% of data that that we work with. They've you've never been able to ask all of that data a question and get an answer back. But AI and AI agents in particular let you finally do that. And so um so I think what's exciting for us is customers are starting to recognize all of the value that they're sitting on that now for the first time ever they can finally tap into. So that's actually kind of the the recognition of that is actually increasing and we're finding that customers are are more and more you know excited by that and and learning much more about what that could mean for their business. >> Okay. And uh if you could paint a day in the life of say a sales rep or VP of marketing something like this, how what would box a uh box agents do? >> Yeah. >> Overnight or something like this that changes how they you know perform their work next day. >> Yeah. So for for a sales rep, let's say um couple quick examples. uh you know if if uh you know one of the biggest challenges that every sales rep deals with is there's a bunch of best practices inside their organization and they usually have to go talk to an expert in the company to get an answer about how they should do a certain task. Um and so what what we have though is we have all of this information in in an organization. We have white papers, we have marketing assets, we have sales materials that actually have the answers to those sales reps questions. So, what if for the first time ever, you could just ask a question of all of this data and get an answer back? That's now possible with with AI agents. So, we see a lot of customers that are collecting all this information. They're creating knowledge hubs that that let them then really turn all that uh unstructured data into real intelligence and business value. So, that's a big use case. We're seeing a lot of people build custom agents to automate certain parts of a of a sales rep's workflow. So, you know, one one thing that you're often doing as a sales rep is you're saying, "Okay, here are the 50 customers that I have to go reach out to, and I want to be able to have personalized materials or personalized sales messaging for each of those customers." So, you can build a custom agent within Box that takes those 50 company names and has a special message for each of them and then produces that as content that then you can go and and do your your pitches with. So we're seeing AI really just accelerate the daily workflows of again almost every knowledge worker. It could be in sales, could be in marketing for collateral generation, it could be in product management for coming up and helping you know create uh product specifications for new features. Uh in engineering, it can help you review your code um or even generate new code. So every job function we we we see as being you know kind of aided and um uh and and really kind of enabled by AI and AI agents. >> And so in your platform there's part of the feature set is that you can create your own agents somehow with some sort of easy functionality. Is that >> Yeah. So we make it we make it very easy to to build AI agents within box. Um, you know, you could we could create a a sales rep agent in, you know, less than two minutes if we wanted with that is sort of, you know, very tuned to our particular sales workflow and our data set. So, customers are able to kind of generate these agents uh rapidly and then boost productivity as a result. >> That's pretty cool. Um, and you've obviously taken uh sort of model agnostic approach. So, use OpenAI, Anthropic, Google. How do you decide which model handles which task? Is there a sort of secret to that or >> Yeah. Um well uh I'd say a limited secret. So the what we do is we we do eval what what their capabilities are. So we we know what models are better at data extraction. We know what models are better for searching for information. We know what models are better at summarizing or answering questions on data. And then within our product, we have some default experiences that align those models with um with the relevant use cases that a customer has. And then we let customers go in and build custom agents that override those model choices. So we will always let you select your own model for a use case. But we do have some defaults that are based on our benchmarking and evaluations that we do. >> Okay. And what sort of controls you how do you put controls in place for you know agents and how can they fail safely? >> Yeah. Um so so for the most part all agents that we have adhere to the security and governance controls that are already in box. So one of the benefits of having your data in box is that you set very clear access permissions and access controls uh for your data. So if you and I are sharing documents, um I'm, you know, making those documents accessible to you. I can always revoke that access. And so if you have an agent that's created, that agent can only access the data that you've that I've shared with you or that you've shared with me if it's if it's your data. And so it it an AI agent within Box can't go off and find other information and answer questions about data you don't have access to. >> So it really adheres to those controls. And obviously over time uh we're going to give customers more and more kind of granular controls on what what do you want agents to be able to do on your data. Maybe some customers will want agents to be able to generate new data or move content around inside of folders and other people maybe won't want that. And so we're going to give you that flexibility in our platform. >> Okay. And you also shipped a uh remote MCP server um chat GPT connector and an integration with Mistl's AI's uh le chat. What does agent agent interop interoperability uh unlock that point integrations couldn't? >> Yeah. So you know the the point integration challenge is is that if you imagine every single AI system has to write to or learn about the APIs of every data system then you just have this massive math problem. Let's say there's a hundred or a thousand you know AI agent companies and then there's 10,000 software companies that have data you know you're in billions and billions of permutations or more of of how to how do you have to kind of wire up all those technologies. So, so what we what we are doing with uh with box is we make it very easy to create those agents um you know within box and then we make it easy for any external system to go and and leverage the data within our platform. Uh and so what MCP does is it lets an external a agentic system or AI system talk to the data in box without ever having seen our API before. So our MCP server effectively helps that AI system understand what tools to use within the Box environment. So we have hundreds or thousands of APIs within Box, but only maybe 10 or 20 are relevant to the common actions that an agent is going to do. And so we we created an MCP server that effectively tells that agent what what tools to use from Box and and how to interact with our system and how to authenticate with it in a very seamless way. That's the power of MCP. So, we're seeing more and more developers leverage MCP as a way of connecting into these environments. >> And what's been more uh popular, the MCP approach into box or are people using, you know, just straight box um in from a dashboard, you know? Yeah. Sort of things. >> Yeah. Well, uh our MCP server is very new. It's it's less than two months old. So, um so, you know, kind of our API integrations is probably the more prevalent use case so far. Yeah. uh but MCP I think will will grow you know quite rapidly because it's increasingly becoming the standard for any AI developer to connect to a different system. So that will will certainly increase over time for sure. >> And what's the agent capability that you haven't shipped yet that you most want to exist? Well, the the dream state is um uh is, you know, what if you could really kind of give an agent any type of task that you need to be able to do with all of your data inbox or or an entire workflow or project. So, we can't yet do this because there's a number of ways where the agent will kind of run out of memory or working memory in in the form of a context window. Um, and it'll it'll sort of fail out at certain points. But the task you want to be able to give is say, "I'd like you to go analyze this company that I'm doing due diligence on, and I want you to take these, you know, 50 documents and I want you to generate a full due diligence report inside of a due diligence sort of shared workspace, and then invite these people into it." >> And it would be great if you could go and kick off that task and just know that with a high degree of of, you know, likelihood that it'll be done effectively and then you go take that data, maybe you share it with people and review it. That's kind of the dream state with AI agents and we imagine that that will be what the future of of a lot of knowledge work looks like. We're very early right now so we can't quite do all of that workflow yet. Uh but it's coming. It's uh it's very much uh you know something that's increasingly plausible and and possible with box >> and so the main constraint for that is the context window at the moment. Is that what holds it back? You just can't get enough information in there for it to make the >> I think you have a you have a mix of factors. So you have context window sort of as the as you use more of the context window the quality of the kind of interaction goes down. So even if a context window says it has a million tokens you can't use all the all million tokens in the exact same way as if you had only had a thousand tokens. >> So you have a context window issue as you scale up. Um you have a capability of the model issue which is does the model know to do the right thing with each of the requests that the user has. And so this is this open question of you know do you start to run into an issue of of the agent starts to perform the wrong actions because its capabilities are are you know not as as um you know not not as uh uh skillful at you know uh processing a document or doing a search or using some kind of tool. So we need capabilities to go up within the models and we need context window to go up in the models. If we can solve those two things, we can then basically begin to operate, you know, any kind of enterprise software system and be able to almost automate any kind of workflow in an enterprise. >> That's crazy, isn't it? And um do you have what about like validating do you have steps which validate that the agent is performing a task in the right way always has come to the right conclusion in the task. >> Yeah. So we we do our own internal kind of eval on on certain kind of data workflows and um you know the only challenge we actually have is that we'll we'll create an eval and then agents will saturate that eval as in they'll you know most of our for our evals that we had two years ago most AI systems can perform them at you know let's say 90% quality. So we have to now actually increase the the the difficulty of those evals to ensure that we're tracking with with the latest capabilities. Um but yes, we're going to eval every new capability that emerges within AI and then figure out, you know, how how effective is it at working with um the latest models and agent capabilities on your data. >> Cool. And um I had a question here. I just wanted to talk about the um AI market more generally. Um, I think this has actually surpassed this recently, but the top six public companies in the US have a combined market cap of over 20 trillion dollars, which is incredible. Uh, larger than the GDP of China, which is the world's second largest economy. >> Yeah. >> Shout out to Northman Trader, uh, who who showed me that the other day. Um, so Nvidia is obviously printing money, selling chips. Microsoft and Google are burning tens of billions a quarter building infrastructure and models. Uh, Box sits on top of their tech to solve real problems. Where does the real money get made long term from your point of view? Is it the companies building AI or the companies actually using it? >> You know, I I think I'm I'm an optimist on this front. I actually think kind of all layers of the stack end up end up benefiting in some way. Um, you know, clearly right now pole position is is Nvidia. So you want to be you know you want to be Nvidia over the past five years uh because you've had you've had control of of the underlying margin structure of of AI effectively and you you've had the raw material that everything needs. So um so I you know you want to be Nvidia start you know starting out uh I think being the hyperscalers that deploy this the the you know the chips and have you know kind of you know relatively um uh you know um relatively kind of um a relative stronghold on that set of chip uh supply chain I think ends up being very helpful for the hyperscalers because there there's only you know three or five companies that really can scale out that that set of chip um uh infrastructure. So you want to be them. And then for the application providers, actually the model providers are obviously, you know, doing well. They're burning money because they're in many cases just in scale up mode. But but I think there's unquestionably you'd want to be open AI and entropic etc. of the past couple years. And then I think now it's the application layer because if you just you kind of just see that flow through you have to get the chips you know created. They have to get into the data centers. The models have to use them. So that kind of represents the first kind of act of the infrastructure buildout. Now at the application layer, we're finally at the point where enterprises can utilize AI across their business processes. So I think you're going to see that now value flow through at the software layer. And you know whether that flows into Salesforce and Service Now and Workday or Box or goes to startups that are brand new like a cursor or something like that, you're going to see I think value creation happen really kind of across the ecosystem. But now I think is the point when you're going to see a lot of that value get generated at the application layer. >> Yeah. And how how quickly do you think um these agents and you know improving the the workflows the the AI is going to permeate through the biggest companies enterprises? Are we going to see big changes over the next year like that are going to influence how businesses operate? Um and when you say um like the end corporations that are maybe not that are not in AI, is that what you're asking? >> Yeah. So enterprises using software like bots. >> Yeah. I think it's going to you know first of all I think we have to be um somewhat realistic that it's it's going to be a multi-year journey. The change management in a large enterprise is is non-trivial. Getting um you know people to use a new set of tools in a new behavior. you know the if you think about let's say there's sort of a two by two of behavior change and tool change that that occurs when you're when you're adopting something um for a lot of the past 10 to 20 years of software we didn't really have behavior change we just had tool change so I upgraded from a chat system to Slack I upgraded from you know Word to Google Docs I upgraded from on premises files to Box so that was a a tool change what we have here is is behavior change which is the way I actually interact with software is totally different. Um the the I'm no longer uh doing all of the work myself within software. I'm actually uh interacting with a effectively AI labor on the other end which is a total behavior change of of of you know how do I actually leverage software. So you have tool and behavior change that represents obviously a much more transformational thing that is going to occur because of of the behavior shift. But it does mean that it's it's it can take longer and it can be harder to go and and actually execute that which means there's more change management. There's more uh that that you have to re-engineer your workflows to get the benefit of agents. Um uh you have to retool some of the ways that you're operating. That that does matter a lot. >> Yeah. Yeah. It feels like um Box is extremely well positioned to uh because it has access to um all this data that other platforms don't have access to because you control that that data side. There's a huge increase in value to enterprises from the advances in AI that they can utilize this data in a much more effective way and nobody else or very few other people have direct access to it. >> Yeah. So, so you know, fortunately, we've spent now about two decades building and earning the trust of enterprises to manage their most important unstructured data. And if you think about all of the things that go into managing data in an enterprise, you need security, compliance, governance, you know, you need to be able to you have to have a lot of scalability. We have to have open APIs. We have to integrate with all of our customers data environments. So, um, so that's what we spent now nearly two decades doing. that that is the the price of entry into being able to do AI on data and and so we're in a position now where we benefit from all of that work that we've done and now we can actually get more of these use cases on your information as a result of all the work that we put in over the years to be able to work with this information. So I I think it is a a big moment for being able to, you know, use your data and thus I think a big a big moment for us. >> And um just touching on some of the financials, I mean, correct me if I'm wrong, but I think Q2 revenue is up 9%. Um RPO, so that sort of represents future revenue up 16%. Um can you just decode that for us? Where's the increase come from? Is it se growth, longer terms, AI related deals? Yeah, I think we're we're definitely seeing a lot of the benefit of AI right now where customers are saying, "I need to have a modern platform to be able to deploy agents on my enterprise information. I need a a platform I trust. I need a platform that's secure. I needed to be able to work with all of my unstructured data." And um uh and so that that's causing customers to now really drive an upgrade cycle into our most advanced features and uh our most advanced capabilities. Um, we have a new plan called Enterprise Advanced and it it, you know, in it has all of the core functionality of our agent builder. It has something called Box apps, so you can build custom applications. We launched a new capability called Box Automate and Box Extract. So, you'll be able to automate workflows with your unstructured data and do data extraction. Both of those become available in Enterprise Advanced. So, that's really causing the recent upswing on the revenue growth and we're super excited. Obviously, we're going to keep just doubling down on on those initiatives. >> And um just a couple more questions, Aaron. Um we've got one. So, net retention was 103% last quarter, which obviously represents um as possible because you're increasing the value of clients you've already got. That's why I can get >> How is Box able to I mean, it's incredible numbers. How's How is Box able to achieve that? What's the one number one thing you think that helps you achieve such high numbers in retention? Yeah, I I do think it's the stickiness of the platform and and the stickiness of these use cases. So, I think by virtue of again being able to leverage AI on your unstructured data, it's causing our customers to have more advanced use cases on our system, which is which is great. Um, that's going to obviously increase stickiness over time and allow our customers to do even more with our our functionality. Mhm. And um why does Box have the structural advantage to become a 10 2030 billion company over the next few years? >> Yeah, I I think it's really because um without commenting on any specific valuation, I think the the structural advantage we have in general that that we think will uh drive you know continued results is um we are living in an era where your data is the most important context for AI. So AI agents need to know about your workflow and your business process. It needs to know about your business and how your enterprise functions. Um and so to get that AI to agents, you need a set of systems and tools that make that very easy to do and make it secure and make it well-governed. And so we're in a position where we're one of the only platforms that can do that at scale with all of this critical business content that drives how agents operate. So that's that's I think why we're in a very strong position and we are we are cranking we are executing at the rate of a startup right now at a faster pace than I've ever seen as a company and we are you know we're grinding 247 on making sure we have all of the most updated functionality for our customers. We live and breathe and and sleep AI. We're you know fortunately right in the middle of Silicon Valley. Um so we get to kind of work with all the AI vendors. We get to see what what's happening across the industry. Um, it's the most fun we've ever had as a company and and I think it's the most important, you know, moment for again unstructured data and what now companies can do with it in the history of computing. >> Yeah, it's a very interesting time to be alive, I think. And um, >> yeah, thank you so much for your time, Aron. I know you're a busy guy, so it's been great to to chat again and also touch on a lot of these um new advances in technology at Box. Um, >> absolutely. Yeah, we I totally appreciate the time. >> And for just before we go for our community, if you had one skill for them to learn this quarter, what that would that be? >> Um, I would honestly just spend as much time with these AI tools as possible. Uh, you know, you're going to you're going to learn so much more by using these technologies versus sort of hypothesizing about them. So if you you know just spending more time in in AI pushing the limits of what these kind of AI agents and AI platforms can do I think will will very much be a very um uh uh you know it'll it'll it'll both prepare you from either a career standpoint or or you know you know trading standpoint or just understanding the world of AI. Um but again you'll be you'll you'll have such a deep understanding of this ecosystem and industry far more than any again you know uh other other way to get that information. >> Cheers. Well it's been a great time. >> Thanks man. Appreciate it.
Box CEO Aaron Levie: AI Agents Will Reshape Work
Summary
Transcript
So one of the biggest use cases that now we're seeing is customers saying okay I have a million contracts a million invoices and I want agents to go and automate the workflows around that information. So we are executing at the rate of a startup right now at a faster pace than I've ever seen as a company and we are you know we're grinding 24/7. We live and breathe and and sleep AI. Um it's the most fun we've ever had as a company and and I think it's the most important moment for unstructured data and what now companies can do with it in the history of computing. Hello everyone, welcome to another session of Opto sessions. Aaron Levy is back. Since the last chat, Box Works just dropped a new wave of launches. Box Extract, Box Automate, and Shield Pro. And Box's AI agents are now working across Microsoft Enthropic, OpenAI to turn unstructured content into actual outcomes. Aaron, welcome back. >> Hey, thanks. Uh, good to uh good to be on again. >> And are you calling calling from HQ today? Uh yes. Yeah. Here in uh lovely California. >> Nice. Um so yeah, eight months ago you said agents would move from demos to dollars. Uh coming out of Box Works, what's the best customer outcome you've seen from your AI agents? >> Yeah. So for us um we're you know we live in a world where um enterprises have just a tremendous amount of of corporate data in the form of contracts and marketing assets and financial documents and invoices and all of that data contains incredibly rich business information in them. Um but unless a person is sort of on a computer looking at a document you're never really like pulling out the real insights from that data. And so what we've built is a set of tools um that let you take AI agents and have those agents process that information and data at scale. So read through a document and extract kind of critical data points from from that document. So in a contract it might be financial information. In an invoice, it might be, you know, who has to, you know, pay you, you know, the the uh the the um uh you know, the the payment and what's the amount owed. In a um research document, it might be all of the key terms uh that relate to that field of research. And so then later you can query it and synthesize that data. So all of our content has this rich information in it and we've never been able to tap into it. And so AI agents finally let us unlock that. So one of the biggest use cases that now we're seeing is customers saying okay I have a million contracts a million invoices a million research you know documents and I want agents to go and automate the workflows around that information. So this has been a uh really the first year where that became available within our platform. And as you mentioned at Box Works, our customer conference, we announced a new capability called Box Extract that makes it very easy to use extraction agents at scale across your data to pull out all of the most relevant insights from that information. >> Okay. And who so who are the users do you think are going to get the most value out of this? Are we talking about like you mentioned like lawyers that are looking through contracts? >> So So I would say kind of quite literally every every job function in an enterprise, you know, deals with with unstructured corporate data. Um, you know, some more than others, but but every single job you're you're looking at some form of a document or a marketing asset or a PowerPoint or a PDF every single day of work. And so agents that can now bring automation to that type of work mean that every single worker will be in some way aided and and enabled uh to have better productivity because of agents and because of uh agents on their data. >> And so this intelligent content management obviously is a big part of your strategy going forward. Um what's the most misunderstood part of this strategy would you say? Um, I think the, you know, fortunately there's actually a growing understanding of of the strategy and what we're up to. We are incredibly excited about, you know, how customers are recognizing the real value of of this information. So, we have more and more companies every single day coming to us saying, you know, I have all of this data that we generate and I I you know, if you think about any average enterprise, about 90% of their information, 90% of their data is unstructured. So think about structured and unstructured as structured data goes into a database. It's a it goes into a CRM system or an HR system. Unstructured data is contracts and documents and marketing assets. It's all it's all sort of messy mess messy, you know, kind of more human readable data. And computers in the past have never been able to really process the 90% of data that that we work with. They've you've never been able to ask all of that data a question and get an answer back. But AI and AI agents in particular let you finally do that. And so um so I think what's exciting for us is customers are starting to recognize all of the value that they're sitting on that now for the first time ever they can finally tap into. So that's actually kind of the the recognition of that is actually increasing and we're finding that customers are are more and more you know excited by that and and learning much more about what that could mean for their business. >> Okay. And uh if you could paint a day in the life of say a sales rep or VP of marketing something like this, how what would box a uh box agents do? >> Yeah. >> Overnight or something like this that changes how they you know perform their work next day. >> Yeah. So for for a sales rep, let's say um couple quick examples. uh you know if if uh you know one of the biggest challenges that every sales rep deals with is there's a bunch of best practices inside their organization and they usually have to go talk to an expert in the company to get an answer about how they should do a certain task. Um and so what what we have though is we have all of this information in in an organization. We have white papers, we have marketing assets, we have sales materials that actually have the answers to those sales reps questions. So, what if for the first time ever, you could just ask a question of all of this data and get an answer back? That's now possible with with AI agents. So, we see a lot of customers that are collecting all this information. They're creating knowledge hubs that that let them then really turn all that uh unstructured data into real intelligence and business value. So, that's a big use case. We're seeing a lot of people build custom agents to automate certain parts of a of a sales rep's workflow. So, you know, one one thing that you're often doing as a sales rep is you're saying, "Okay, here are the 50 customers that I have to go reach out to, and I want to be able to have personalized materials or personalized sales messaging for each of those customers." So, you can build a custom agent within Box that takes those 50 company names and has a special message for each of them and then produces that as content that then you can go and and do your your pitches with. So we're seeing AI really just accelerate the daily workflows of again almost every knowledge worker. It could be in sales, could be in marketing for collateral generation, it could be in product management for coming up and helping you know create uh product specifications for new features. Uh in engineering, it can help you review your code um or even generate new code. So every job function we we we see as being you know kind of aided and um uh and and really kind of enabled by AI and AI agents. >> And so in your platform there's part of the feature set is that you can create your own agents somehow with some sort of easy functionality. Is that >> Yeah. So we make it we make it very easy to to build AI agents within box. Um, you know, you could we could create a a sales rep agent in, you know, less than two minutes if we wanted with that is sort of, you know, very tuned to our particular sales workflow and our data set. So, customers are able to kind of generate these agents uh rapidly and then boost productivity as a result. >> That's pretty cool. Um, and you've obviously taken uh sort of model agnostic approach. So, use OpenAI, Anthropic, Google. How do you decide which model handles which task? Is there a sort of secret to that or >> Yeah. Um well uh I'd say a limited secret. So the what we do is we we do eval what what their capabilities are. So we we know what models are better at data extraction. We know what models are better for searching for information. We know what models are better at summarizing or answering questions on data. And then within our product, we have some default experiences that align those models with um with the relevant use cases that a customer has. And then we let customers go in and build custom agents that override those model choices. So we will always let you select your own model for a use case. But we do have some defaults that are based on our benchmarking and evaluations that we do. >> Okay. And what sort of controls you how do you put controls in place for you know agents and how can they fail safely? >> Yeah. Um so so for the most part all agents that we have adhere to the security and governance controls that are already in box. So one of the benefits of having your data in box is that you set very clear access permissions and access controls uh for your data. So if you and I are sharing documents, um I'm, you know, making those documents accessible to you. I can always revoke that access. And so if you have an agent that's created, that agent can only access the data that you've that I've shared with you or that you've shared with me if it's if it's your data. And so it it an AI agent within Box can't go off and find other information and answer questions about data you don't have access to. >> So it really adheres to those controls. And obviously over time uh we're going to give customers more and more kind of granular controls on what what do you want agents to be able to do on your data. Maybe some customers will want agents to be able to generate new data or move content around inside of folders and other people maybe won't want that. And so we're going to give you that flexibility in our platform. >> Okay. And you also shipped a uh remote MCP server um chat GPT connector and an integration with Mistl's AI's uh le chat. What does agent agent interop interoperability uh unlock that point integrations couldn't? >> Yeah. So you know the the point integration challenge is is that if you imagine every single AI system has to write to or learn about the APIs of every data system then you just have this massive math problem. Let's say there's a hundred or a thousand you know AI agent companies and then there's 10,000 software companies that have data you know you're in billions and billions of permutations or more of of how to how do you have to kind of wire up all those technologies. So, so what we what we are doing with uh with box is we make it very easy to create those agents um you know within box and then we make it easy for any external system to go and and leverage the data within our platform. Uh and so what MCP does is it lets an external a agentic system or AI system talk to the data in box without ever having seen our API before. So our MCP server effectively helps that AI system understand what tools to use within the Box environment. So we have hundreds or thousands of APIs within Box, but only maybe 10 or 20 are relevant to the common actions that an agent is going to do. And so we we created an MCP server that effectively tells that agent what what tools to use from Box and and how to interact with our system and how to authenticate with it in a very seamless way. That's the power of MCP. So, we're seeing more and more developers leverage MCP as a way of connecting into these environments. >> And what's been more uh popular, the MCP approach into box or are people using, you know, just straight box um in from a dashboard, you know? Yeah. Sort of things. >> Yeah. Well, uh our MCP server is very new. It's it's less than two months old. So, um so, you know, kind of our API integrations is probably the more prevalent use case so far. Yeah. uh but MCP I think will will grow you know quite rapidly because it's increasingly becoming the standard for any AI developer to connect to a different system. So that will will certainly increase over time for sure. >> And what's the agent capability that you haven't shipped yet that you most want to exist? Well, the the dream state is um uh is, you know, what if you could really kind of give an agent any type of task that you need to be able to do with all of your data inbox or or an entire workflow or project. So, we can't yet do this because there's a number of ways where the agent will kind of run out of memory or working memory in in the form of a context window. Um, and it'll it'll sort of fail out at certain points. But the task you want to be able to give is say, "I'd like you to go analyze this company that I'm doing due diligence on, and I want you to take these, you know, 50 documents and I want you to generate a full due diligence report inside of a due diligence sort of shared workspace, and then invite these people into it." >> And it would be great if you could go and kick off that task and just know that with a high degree of of, you know, likelihood that it'll be done effectively and then you go take that data, maybe you share it with people and review it. That's kind of the dream state with AI agents and we imagine that that will be what the future of of a lot of knowledge work looks like. We're very early right now so we can't quite do all of that workflow yet. Uh but it's coming. It's uh it's very much uh you know something that's increasingly plausible and and possible with box >> and so the main constraint for that is the context window at the moment. Is that what holds it back? You just can't get enough information in there for it to make the >> I think you have a you have a mix of factors. So you have context window sort of as the as you use more of the context window the quality of the kind of interaction goes down. So even if a context window says it has a million tokens you can't use all the all million tokens in the exact same way as if you had only had a thousand tokens. >> So you have a context window issue as you scale up. Um you have a capability of the model issue which is does the model know to do the right thing with each of the requests that the user has. And so this is this open question of you know do you start to run into an issue of of the agent starts to perform the wrong actions because its capabilities are are you know not as as um you know not not as uh uh skillful at you know uh processing a document or doing a search or using some kind of tool. So we need capabilities to go up within the models and we need context window to go up in the models. If we can solve those two things, we can then basically begin to operate, you know, any kind of enterprise software system and be able to almost automate any kind of workflow in an enterprise. >> That's crazy, isn't it? And um do you have what about like validating do you have steps which validate that the agent is performing a task in the right way always has come to the right conclusion in the task. >> Yeah. So we we do our own internal kind of eval on on certain kind of data workflows and um you know the only challenge we actually have is that we'll we'll create an eval and then agents will saturate that eval as in they'll you know most of our for our evals that we had two years ago most AI systems can perform them at you know let's say 90% quality. So we have to now actually increase the the the difficulty of those evals to ensure that we're tracking with with the latest capabilities. Um but yes, we're going to eval every new capability that emerges within AI and then figure out, you know, how how effective is it at working with um the latest models and agent capabilities on your data. >> Cool. And um I had a question here. I just wanted to talk about the um AI market more generally. Um, I think this has actually surpassed this recently, but the top six public companies in the US have a combined market cap of over 20 trillion dollars, which is incredible. Uh, larger than the GDP of China, which is the world's second largest economy. >> Yeah. >> Shout out to Northman Trader, uh, who who showed me that the other day. Um, so Nvidia is obviously printing money, selling chips. Microsoft and Google are burning tens of billions a quarter building infrastructure and models. Uh, Box sits on top of their tech to solve real problems. Where does the real money get made long term from your point of view? Is it the companies building AI or the companies actually using it? >> You know, I I think I'm I'm an optimist on this front. I actually think kind of all layers of the stack end up end up benefiting in some way. Um, you know, clearly right now pole position is is Nvidia. So you want to be you know you want to be Nvidia over the past five years uh because you've had you've had control of of the underlying margin structure of of AI effectively and you you've had the raw material that everything needs. So um so I you know you want to be Nvidia start you know starting out uh I think being the hyperscalers that deploy this the the you know the chips and have you know kind of you know relatively um uh you know um relatively kind of um a relative stronghold on that set of chip uh supply chain I think ends up being very helpful for the hyperscalers because there there's only you know three or five companies that really can scale out that that set of chip um uh infrastructure. So you want to be them. And then for the application providers, actually the model providers are obviously, you know, doing well. They're burning money because they're in many cases just in scale up mode. But but I think there's unquestionably you'd want to be open AI and entropic etc. of the past couple years. And then I think now it's the application layer because if you just you kind of just see that flow through you have to get the chips you know created. They have to get into the data centers. The models have to use them. So that kind of represents the first kind of act of the infrastructure buildout. Now at the application layer, we're finally at the point where enterprises can utilize AI across their business processes. So I think you're going to see that now value flow through at the software layer. And you know whether that flows into Salesforce and Service Now and Workday or Box or goes to startups that are brand new like a cursor or something like that, you're going to see I think value creation happen really kind of across the ecosystem. But now I think is the point when you're going to see a lot of that value get generated at the application layer. >> Yeah. And how how quickly do you think um these agents and you know improving the the workflows the the AI is going to permeate through the biggest companies enterprises? Are we going to see big changes over the next year like that are going to influence how businesses operate? Um and when you say um like the end corporations that are maybe not that are not in AI, is that what you're asking? >> Yeah. So enterprises using software like bots. >> Yeah. I think it's going to you know first of all I think we have to be um somewhat realistic that it's it's going to be a multi-year journey. The change management in a large enterprise is is non-trivial. Getting um you know people to use a new set of tools in a new behavior. you know the if you think about let's say there's sort of a two by two of behavior change and tool change that that occurs when you're when you're adopting something um for a lot of the past 10 to 20 years of software we didn't really have behavior change we just had tool change so I upgraded from a chat system to Slack I upgraded from you know Word to Google Docs I upgraded from on premises files to Box so that was a a tool change what we have here is is behavior change which is the way I actually interact with software is totally different. Um the the I'm no longer uh doing all of the work myself within software. I'm actually uh interacting with a effectively AI labor on the other end which is a total behavior change of of of you know how do I actually leverage software. So you have tool and behavior change that represents obviously a much more transformational thing that is going to occur because of of the behavior shift. But it does mean that it's it's it can take longer and it can be harder to go and and actually execute that which means there's more change management. There's more uh that that you have to re-engineer your workflows to get the benefit of agents. Um uh you have to retool some of the ways that you're operating. That that does matter a lot. >> Yeah. Yeah. It feels like um Box is extremely well positioned to uh because it has access to um all this data that other platforms don't have access to because you control that that data side. There's a huge increase in value to enterprises from the advances in AI that they can utilize this data in a much more effective way and nobody else or very few other people have direct access to it. >> Yeah. So, so you know, fortunately, we've spent now about two decades building and earning the trust of enterprises to manage their most important unstructured data. And if you think about all of the things that go into managing data in an enterprise, you need security, compliance, governance, you know, you need to be able to you have to have a lot of scalability. We have to have open APIs. We have to integrate with all of our customers data environments. So, um, so that's what we spent now nearly two decades doing. that that is the the price of entry into being able to do AI on data and and so we're in a position now where we benefit from all of that work that we've done and now we can actually get more of these use cases on your information as a result of all the work that we put in over the years to be able to work with this information. So I I think it is a a big moment for being able to, you know, use your data and thus I think a big a big moment for us. >> And um just touching on some of the financials, I mean, correct me if I'm wrong, but I think Q2 revenue is up 9%. Um RPO, so that sort of represents future revenue up 16%. Um can you just decode that for us? Where's the increase come from? Is it se growth, longer terms, AI related deals? Yeah, I think we're we're definitely seeing a lot of the benefit of AI right now where customers are saying, "I need to have a modern platform to be able to deploy agents on my enterprise information. I need a a platform I trust. I need a platform that's secure. I needed to be able to work with all of my unstructured data." And um uh and so that that's causing customers to now really drive an upgrade cycle into our most advanced features and uh our most advanced capabilities. Um, we have a new plan called Enterprise Advanced and it it, you know, in it has all of the core functionality of our agent builder. It has something called Box apps, so you can build custom applications. We launched a new capability called Box Automate and Box Extract. So, you'll be able to automate workflows with your unstructured data and do data extraction. Both of those become available in Enterprise Advanced. So, that's really causing the recent upswing on the revenue growth and we're super excited. Obviously, we're going to keep just doubling down on on those initiatives. >> And um just a couple more questions, Aaron. Um we've got one. So, net retention was 103% last quarter, which obviously represents um as possible because you're increasing the value of clients you've already got. That's why I can get >> How is Box able to I mean, it's incredible numbers. How's How is Box able to achieve that? What's the one number one thing you think that helps you achieve such high numbers in retention? Yeah, I I do think it's the stickiness of the platform and and the stickiness of these use cases. So, I think by virtue of again being able to leverage AI on your unstructured data, it's causing our customers to have more advanced use cases on our system, which is which is great. Um, that's going to obviously increase stickiness over time and allow our customers to do even more with our our functionality. Mhm. And um why does Box have the structural advantage to become a 10 2030 billion company over the next few years? >> Yeah, I I think it's really because um without commenting on any specific valuation, I think the the structural advantage we have in general that that we think will uh drive you know continued results is um we are living in an era where your data is the most important context for AI. So AI agents need to know about your workflow and your business process. It needs to know about your business and how your enterprise functions. Um and so to get that AI to agents, you need a set of systems and tools that make that very easy to do and make it secure and make it well-governed. And so we're in a position where we're one of the only platforms that can do that at scale with all of this critical business content that drives how agents operate. So that's that's I think why we're in a very strong position and we are we are cranking we are executing at the rate of a startup right now at a faster pace than I've ever seen as a company and we are you know we're grinding 247 on making sure we have all of the most updated functionality for our customers. We live and breathe and and sleep AI. We're you know fortunately right in the middle of Silicon Valley. Um so we get to kind of work with all the AI vendors. We get to see what what's happening across the industry. Um, it's the most fun we've ever had as a company and and I think it's the most important, you know, moment for again unstructured data and what now companies can do with it in the history of computing. >> Yeah, it's a very interesting time to be alive, I think. And um, >> yeah, thank you so much for your time, Aron. I know you're a busy guy, so it's been great to to chat again and also touch on a lot of these um new advances in technology at Box. Um, >> absolutely. Yeah, we I totally appreciate the time. >> And for just before we go for our community, if you had one skill for them to learn this quarter, what that would that be? >> Um, I would honestly just spend as much time with these AI tools as possible. Uh, you know, you're going to you're going to learn so much more by using these technologies versus sort of hypothesizing about them. So if you you know just spending more time in in AI pushing the limits of what these kind of AI agents and AI platforms can do I think will will very much be a very um uh uh you know it'll it'll it'll both prepare you from either a career standpoint or or you know you know trading standpoint or just understanding the world of AI. Um but again you'll be you'll you'll have such a deep understanding of this ecosystem and industry far more than any again you know uh other other way to get that information. >> Cheers. Well it's been a great time. >> Thanks man. Appreciate it.