AI agents in enterprise environments require access to unstructured business content (approximately 90% of corporate data) to be effective, as agents trained only on public internet data lack organization-specific context. The fragmentation of enterprise content across multiple systems creates significant challenges for agent deployment, including outdated information retrieval, access control inconsistencies, and security vulnerabilities. Organizations must implement secure content management platforms that connect AI agents to enterprise data while maintaining governance controls, enabling agents to perform complex workflows like contract analysis, document processing, and business intelligence extraction at scale.
Installieren Sie unsere Erweiterung an, um sofort in jedem Video zu suchen
Keynote: Unlocking the power of AI begins with your contentHinzugefügt:
Hello, I'm Aaron Levy, co-founder and CEO of Box, and welcome to our virtual summit. Today, you're going to hear from our product leadership team around the future of the Box platform and how we're connecting the full power of AI agents with enterprise content securely.
Also, you're going to hear from amazing customers that are pushing the boundaries of the Box platform, as well as critical partners that we're working with to bring the full power of AI to your organization. There's also a series of deep dives that we'd love for you to participate in that go much deeper in critical areas that you care about.
Let's get the event started. At Box, our mission is to power how the world works together. And we couldn't be more proud than to be able to deliver on this mission with over 120,000 organizations globally. We have the great honor of being able to work with some of the world's fastest growing companies as well as some of the industrial giants that are transforming the way the world works every single day. leading aerospace and energy companies like GE, financial institutions like Morgan Stanley, technology companies like Broadcom. All of these organizations are shaping how our world works and how it operates. And what's incredibly exciting is that we get to see a front row picture of how the world work is shaping through the eyes of these organizations.
And one thing that we all know is that AI is transforming everything about how we work. And we hear this day in and day out from all of our customers. What started out just a few years ago of being able to have AI agents that could answer any kind of question inside of a process. We could chat back and forth has quickly become AI agents that can begin to execute on nearly any type of task that we give them. And those tasks are increasing in scale and volume. So the agents can actually run longer and work on harder and harder problems for us. We're seeing a constant increase in again task complexity that agents can deliver on. But where we're going and what you're going to see from today's keynote is really what happens when you have a fleet of agents or a swarm of agents that can actually automate multiple workflows. We can deploy these agents in a business process and really revolutionize the underlying workflows within our organizations. The simple way that we like to think about it is imagine if every employee in your organization or every workflow in your company had access to an expert analyst or researcher or domain expert in engineering or legal or marketing or sales. What if they'd access that expert individual that could work a thousand times faster? They still have to be given the right task to do. They have to have the right context to work with.
They have to be applied inside of the right guard rails. But what if they could work a thousand times faster? what would we be able to unleash inside of our organization? This is the kind of mental model that we are working with.
Now, the first set of use cases that we're starting to see is how do we accelerate our knowledge work? How do you answer questions faster in an organization? How do you bring expertise to any kind of problem that arises? This could be reviewing a contract or looking at financial information for key insights. How do you look at product materials and get insights about your product roadmap? This is enabling increased productivity for every individual worker in an organization.
But the real impact of AI agents is not just what we do with them as individuals. But when you think about deploying them in a workflow or a business process across the enterprise, we can begin to transform our processes with agentic workflows. This can have a much bigger impact across the organization. Like how do you accelerate client onboarding so you can bring on customers far faster? Being able to automate approvals so we can streamline our projects much more quickly. Being able to personalize marketing in into any segment or region that we want to go and serve. Be able to have faster product development because we can not only generate code faster, but we can look at customer insights and accelerate that side of the pipeline. or ultimately reduce business risk because we can have agents moving across our data landscape discovering what issues we might run into or areas where we need to improve compliance or our security posture. So this is the real impact of AI agents in the enterprise when we actually go transform the underlying workflows in our organizations. Now to transform with AI agents need to know everything about your business. You can think about these agents as these incredible super intelligent machines or systems, but they've been trained on the public internet. They've been trained on large amounts of public information. They don't know about your unique decisions and the unique areas that make your business differentiated and what matters to your organization. So, they're super intelligent systems that need the context of your particular organization to be effective. What do they need access to? Well, they need to know your product specifications. They need to know your research. They need to have access to critical marketing assets that you've created. They need to know your HR policies and best practices and everything else that makes your organization unique. Without that context, the agent is again going to be the same across any organization and not be able to specifically turbocharge your organization or the agent will make the wrong decisions for your company. So all of that context is necessary for an agent to be effective inside of a workflow or to accelerate our knowledge work. And we see that the vast majority of that context lives inside your unique enterprise content. It lives inside the product materials that help us launch new products. It lives inside of our financial records that help us close the books. It lives inside of our marketing assets which help us actually market to customers or have brand guidelines that we stay on top of. It lives inside the financial records that allow us to onboard clients much faster or the sales resources that let a sales rep answer the right question for a customer or the contracts that help us close deals. All of that is the unstructured data or the enterprise content that becomes critical context for AI agents and AI systems. So the real power of AI is when we can successfully and securely connect that enterprise content to our AI agents.
That is how we're ultimately going to be able to get the real gains from AI agents in the enterprise. And what's incredibly exciting for us at Box is actually all that unstructured data is the vast majority of information in the enterprise. It's about 90% of corporate data that we work with. So 10% of our data is information that we've always been able to query and analyze and compute and build dashboards on. But 90% of the data is unstructured. It's stuff inside of PowerPoint presentations or documents or spreadsheets or PDFs or images, videos, and other file types.
All of that unstructured data contains this incredible resource for our organization that we've never been able to tap into. We've never been able to compute and query and analyze all that data. It's been hard for computers to really understand what's inside this information. So, organizations are sitting on 90% of their data that we've never been able to tap into. But what if you could have AI work across this information? What would you be able to glean from your organization? What would you be able to pull out on the decisions that you need to make or the processes that you can go accelerate or the information that becomes useful for agents that are going in automating workflows. Imagine an agent that had access to all of your critical product materials. You'd be able to enable any employee to ask a question like, "Well, what's the risk in this product timeline? What do I need to do to to make sure I shuffle the priorities to make our execution go better? All of that information lives inside your product specs or your product roadmap?
What if you could give an AI agent access to all of your clinical research information securely and in a HIPACO compliant way? What if you could begin to figure out what patients you're having new discoveries from or what in your research is now leading to a new breakthrough? What if you could take financial documents and financial information and when you're going to a client inside of a financial services firm or a wealth management organization, what if you could deliver the right investment advice at the right time based on that particular client's history or new financial products that you have? Or what if you were onboarding a new customer or doing an M&A transaction and you had a bunch of critical data that you want an agent to be able to go and automate the review of? you could figure out new risks or new opportunities or new challenges within that information. Could be contracts or KYC or financial documents inside of areas like an investment bank or law firm. So all of that information is living right now in our unstructured data, but we've never been able to automate any of the work on it before.
But for the first time ever, AI lets us do that. Now there's a catch which is that the way that we are managing our data or our unstructured data in particular in most organizations is fundamentally broken and it won't let us get the real gains from AI. And this is because our content tends to be fragmented across a wide variety of systems. It's in network file shares.
It's in document management tools. It's in collaboration tools. Maybe some of it's in the cloud. But even when it's in the cloud, it could be in fragmented environments or environments that don't work well for getting that information to agents. Now, this type of architecture has already been a problem historically. It means it's hard to get access to the right data that we want to work with. It's often very insecure to have data fragmented across all these systems and it can be extremely costly to have so much redundancy in our infrastructure. So, this has been a pain for organizations for many years, but now it's actually an existential challenge. It's no longer something that we can just sort of get by with and and sort of deal with or manually kind of cover up. This becomes an existential challenge in a world of agents. Now, the reason for this is that agents are going to have a really really hard time working with all of that fragmented infrastructure that we have. The first is that agents will often mistakenly pull content from the wrong source, which means they'll often be working with outof-date information. Imagine how many places you've stored contracts in or marketing assets in or research materials in. Well, every single one of those fragmented silos presents a risk for an agent to pull the wrong piece of data or pull an outdated source of information. So, that becomes a massive challenge for agent productivity because agents won't know which data to work with. The second big challenge is that you have an access control sprawl problem, which is the more systems you have, the more variance you're going to have in access controls, which means that agents in some cases won't have access to the information they need, but maybe more problematically in some systems, they'll have too much access, which means that they'll have an overexposure of information. So the agent might grab a file that they shouldn't be able to access, but because some user did have access to it 5 years ago, now that agent is able to answer a question that they shouldn't be able to.
And then finally, the more fragmentation you have, the harder the challenge is as you have multiple agent systems that you need to be able to work with. Some of those systems won't be compatible with the different content repositories you're using, or it's just a nightmare to be able to manage all of that complexity. So the way that we manage content today is not going to work in a future with AI agents that are running around maybe at a scale of 100 or a thousand times more people than are with an organization. So enterprises need a platform that can connect content to AI securely. We need to be able to securely manage our most important information in the enterprise and we need to be able to connect up AI agents and workflows to that content in a secure fashion. And that's what we're building with intelligent content management. At Box, we're building the leading intelligent content management platform that can securely manage, organize, store, and govern your most important and sensitive enterprise information. And then connect it up with the full stack of agents that you want to be able to work with. Now, at the base layer of our platform, we offer a global infrastructure system that powers unlimited storage for customers. So, you can throw any hard data problem that you want at us. On top of that, we have a layer of data security, compliance, and protection that ensures that you can use Box in your most critical workflows and business processes. Things like automatic threat detection, data classification, data governance, retention management, and much more. All to keep your content safe and secure.
Then we have a layer of content services for things like being able to manage billions of files in an enterprise or be able to attach metadata to any amount of content. The ability to automate workflows or publish content to teams or departments or be able to build no code applications on top of Box. That's our content services layer. Now we've added a new layer which is our AI platform layer. And in our AI platform layer, we have an agentic harness that allows you to do the deepest amount of work that you want in enterprise content with AI agents. We let you develop custom agents in just a matter of seconds. And you can connect with any leading AI model like GPT 5.5, Opus 4.7, Gemini 3.1, and more.
So you can bring the full power of the leading frontier AI models to your enterprise content. But that's not all.
We know that there's enterprise agent systems that customers are deploying from a variety of frontier labs and platforms. So this means that if your enterprise is deploying something like chatbt or codeex or claude co-work or claude or co-pilot co-work box will equally integrate and embed into any of those agentic systems as well. So wherever your users are working from, wherever you're deploying agents, and no matter what application anyone is working within, your content can securely show up in those different systems. So instead of having to fragment your data or store in lots of locations, we give you a single platform that connects your people, agents, and applications to your enterprise content securely.
Now, we know that customers are going on a broad AI transformation journey right now, and we want to help every one of you on this journey. You're going to hear about a number of breakthrough new products that we're delivering to market that will help you on this journey. The first stage of this journey that we tend to see is how do you start to accelerate knowledge work with AI agents? How do we take the expertise that's already within our enterprise content and unleash that for any employee to be able to work with? Whether they're talking to a single file or their entire Box account or specific hubs of data that they want to be able to work with. We want to accelerate the day-to-day work that every one of your employees or people in your organization are doing. Next, given there's so much unstructured data in the enterprise and there's a wealth of information in this data, we want to let you mine intelligence at scale from this enterprise content. This could be things like extracting critical data from contracts or from financial assets. This could be looking at invoices and being able to pull out the critical information from those documents or maybe looking at brand asset imagery and being able to appropriately classify all of that data. So once you can mine intelligence from your enterprise content, you can analyze it, you can query it, you can build dashboards on it, you can better understand what's happening inside your business within your unstructured data. And then finally, once you have that structured data in many cases and you have increasingly more powerful longrunning AI agents, we can begin to transform our processes with agentic workflows. This is why we've launched Box Automate, which lets you build end-to-end workflows directly in Box, both connecting the deterministic capabilities in Box as well as people and AI agents in any kind of business process. And you're going to hear a bit more in a few minutes from the team on what that looks like. So, at Box, we're here to help you with your AI transformation journey. We want to be your partner for enabling you to leverage all of the value and power within your enterprise content to transform the way that you work. And with that, I'm excited to hand it over to Ben Cous, our CTO.
>> Thanks, Aaron. Hi, I'm Ben Cous, CTO of Box, and I'm here to talk about our AI strategy. So, at Box, we've been working on agents for a while, but in the last six months or so, there's been a dramatic increase in the technology that powers these agents that allow for much more complex and much more sophisticated tasks in an enterprise. Previously, agents struggled to do very complex tasks reliably, especially when they had a lot of context. Users weren't quite ready to ask the agents to do complex things because they expected answers in fixed times. And in many cases, it was hard for those agents to get access to data that is important for their job.
Today, the latest models from the frontier vendors are very good with complex reasoning and they can do much more sophisticated tasks. Additionally, the harnesses and the capabilities of the agent and the way the agents are interacting with users are now able to facilitate the latest multi-step longunning tasks to power this more sophisticated type of enterprise knowledge work. The way the agents go and retrieve data from their enterprise systems has been improving. So they're able to get access to more of the most critical data that help them do these more important tasks. We're starting to see a new paradigm emerge where we're working with agents as co-workers. In some ways, we've seen this with the way that the engineering has evolved. And we believe that many of these new capabilities are also coming to general knowledge work where people are starting to ask agents to do more complex tasks.
Agents are creating more information and sharing the data back to the users. And in this paradigm, you end up with a world where you start to treat these agents like co-workers that you're collaborating with as opposed to software that is doing work for you. And in this kind of scenario, you also have the primary way by which you communicate with other users, which is often through things like sharing context via files, via presentations, via documents, via spreadsheets. You're seeing that this is starting to also become the way that you're sharing information with agents and the agents then share the data back to you when they're able to complete their task. With this new paradigm, there's a new challenge that's emerging for enterprises. You need to find a way to be able to protect your data while enabling agents to work on it. And this includes being able to take what is often a massive amount of data that exists in every enterprise in the form of unstructured data and allowing agents to be able to access it. Think about an agent that needs to go through terabytes or pabytes of data that you own. So they can find the most relevant context and able to do this in a secure way that restricts the agent down to making sure that is able to only access the data is authorized for for that type of project that it's working on. This whole time you need to ensure that you have a protecting against the newest generation of attacks including when attackers try to attack the agents specifically to get them to help to leak your data.
Similarly, when you're talking about collaborating with agents, you need to ensure that you have the right ability to share data with agents and have them share data back with you so that you're able to work on the same data and to do it in a way that is secure and is efficient. And for Box, these are the kind of technologies that we have built for the Agentic era. We have a context retrieval system that's able to give agents secure and extensive access to your files so that they're able to find their right data and to handle the kind of challenges like making sure that the files are properly converted to a format that's friendly and that the agent can understand in addition to doing the latest for hybrid search, semantic search combined with lexical search and being able to handle different file types for multimmodal access. And with this, we're able to then also handle the collaboration layer between not just a user-to-user collaboration, but then also the ability to have a userto agent type of collaboration where you're making sure the agent has the right permissions to read the data that you're interested in in addition to making sure that they can share their data back to you. And on top of all of this, applying the right security and governance controls such that your agent is permissions aware. And we're applying protections like prompt injection defense in addition to action guardrails to ensure the agent doesn't do something with your data that you don't want it to. In a box, we have a multi-art platform which is the headless unstructured data platform to be powering all these capabilities from the underlying infrastructure that is global in nature and handles all of the security data protection and compliance details that you need across your users and across your agents onto the type of services that you need from search to query to being able to do things like agents workingflows working across no code apps and then the AI platform layer that then lets you have the ability to have agents that are working across your data using these type of agent guard rails and using the latest agentic approaches so that you're able to apply agents in your organization. And what's very critical is that there are different types of agents here including agents from box and agents from third parties that are cooperating to help you in this case. And so the agents in box will be able to securely access data. If enterprises are building their own agents or if they have agents from OpenAI, agents from Gemini, agents from cloud, then they're able to reach into box via our API, our MCP server, our CLIs, our SDKs, so that they're able to then be able to access your enterprise data in a safe and secure way. And when we talk about what's possible now that wasn't possible before, here are a few examples. Things like if you needed to, let's say, do an analysis of a series of contracts. Maybe you wanted to know how some of the latest changes in the world might affect some of your contracts.
being able to do things like a full liability analysis across thousands or tens of thousands of files. This is the kind of thing that used to take a group of people a very long time to do to go through and analyze many many different files to be able to read through all the contracts can now be done in minutes because an agent would be able to do it.
Also, things like being able to respond to RFPs. So, you have subject matter experts and a knowledge base of material and then you're going to want to be able to take that and then to take in requests from customers to respond to critical questions for your organization and have agents respond to that. Also, another example might be something like being able to update the budget from last year to this year. Being able to take the old budget, being able to take a new strategy, a PowerPoint, maybe a transcript associated with some meetings that you've had, and then be able to actually update the model in Excel.
These are three examples of many, many different examples across industries, across companies where people are using the latest power of AI agents to be able to then deliver on this idea that agents will begin assisting users in the enterprise. With all of these examples and more, the really critical aspects of it are to prepare your data to be able to work with any agents be agents inside a box that we'll tell you about more in a moment or agents that come from third parties that need to access the data inside a box. But for all of them, they have the concerns associated with enterprise scale retrieval. Making sure that security is built in every interaction and ensuring that you have the ability to have universal agent interoperability so that you can use different agents and you're not going to be locked into a full ecosystem of agents permanently because they are controlling your data. The data is available for all the different agents.
And with that, let me turn it over to Matt who will tell you more about our product road map.
>> Thanks, Ben. Hello everyone and welcome to the product keynote. My name is Matt and I lead our AI agents work at Box.
Today we'll show you how Box is using AI to transform knowledge work through intelligent content workflows, automation, and secure enterprise experiences. Let's dive in. So let's start with how Box is accelerating knowledge work for individuals and teams. Knowledge work is at the heart of every organization. It's how our teams research new opportunities, design new products, and make decisions. Every organization, every team even will approach these processes differently.
This is how we differentiate ourselves.
But at their core, they all involve finding, analyzing, and summarizing information to generate new insights and reach decisions. These are all tasks that AI can now make much more efficient. Let's use a compliance questionnaire as an example. You need to read the document, understand what is being asked for, source any missing information, and coordinate the response, collating all of that search for data into what you actually need to draft an actual response, then probably have to review it with team members, and then publish and send the document to where it needs to go. Until recently, AI could help with different steps. You can use QA agents to summarize individual documents or groups of documents in the system it supports, but it couldn't help you with the endtoend process. This is a great improvement for teams, but it still leaves them to do a lot. We realize that to get this type of knowledge work done, you don't want multiple agents. Multiple agents and multiple tools lead to user confusion on what to use and when. That friction can be the difference in adoption in the workplace. Users think in terms of outputs getting work done. What you want is one agent with multiple capabilities.
You need the box agent. The box agent is your new partner to plan and execute these endtoend knowledge work projects.
The box agent can accomplish complex tasks in one session based on its ability to plan the steps to completion and execute them with the tools and capabilities we are giving it. Then the knowledge your teams and their agents produce is stored straight onto Box giving you full control over it. Your AI agents and sessions will also follow you throughout Box so you can continue conversations to get your work done. So, it doesn't matter if you start from a shared link in Box Preview or from a document set in Boxhubs, your agent will be there. Soon, you'll be able to create first class spreadsheets, slide decks, and documents of many different formats just from a conversation with Box AI.
Now, let's see the Box AI agent in action.
>> Here we have logistics, brokerage, warehousing, and trucking contracts stored in Box. When leadership asks about tariff exposure, teams often have to review dozens of agreements to understand where the risk actually sits.
An AI agent can help analyze the documents and surface the key patterns for review. In Box AI, we'll first add our documents as a source. This way, our agent can securely access these contracts as enterprise context. We'll now ask Box AAI to review the contracts, categorize tariff risk, flag higher risk vendors, and prepare a briefing for legal and procurement. We'll go ahead and submit this.
Our agent is going to analyze the agreements and organize the findings into a structured report. It might take a few minutes for the agent to complete, so we'll actually jump ahead to the finished result.
When it's complete, we can check out the report saved back two bucks. Our agent summarizes where terrorisk appears across contracts, highlights vendors that might need closer review, and also suggest areas that legal and procurement teams may want to address.
This is AI agents in action, helping teams analyze enterprise content and service critical risks and insights.
Now that you've seen the Box agent in action, let's take a look at how to build, deploy, and manage agents at scale. With Box AI Studio, teams can optimize agents to work for them. They can create agents for specific use cases, link agents to the right context, and control agent access and behavior.
As we democratize agent creation for permitted end users, teams can safely deploy their own agents, the use of AI capabilities expands and the transformation accelerates. To explore this future, please welcome back Box CEO Aaron Levy for a conversation with OpenAI.
I'm incredibly excited to uh have a longtime partner of Box OpenAI, Dom Grill, who leads technical success at OpenAI. Welcome, Dom. Thank you so much for having me.
>> Thank you so much for uh for being here.
Um we wanted to talk about what the future of agents in the enterprise looks like. Obviously no no better company or person to talk to than what you're seeing through the lens of OpenAI and model progress and then a little bit about what the future of software looks like. So maybe first if we just start of maybe a little bit of lay of the land what's happening in the enterprise with agents um and what's happening with the rate of progress with these AI model capabilities that you guys are delivering. I'd say there's a couple things that are happening is that November there was this seismic shift in what are agents capable of and I think that that was a result of two things.
One the models themselves and the underlying intelligence and reasoning getting really good and then two the agent harnesses getting significantly better. So agents suddenly have access to a lot more tools. They know how to use a computer really good. If you think about what can't be substantiated, what type of knowledge worker tasks can't be substantiated by really good coding, really good use of a computer, um I think that there's not a huge amount of type of tasks that are left that are not going to be untouched or changed by agents. And as a result of that, we're seeing kind of two categories of agents emerge. Um the first I would just call the sort of like AI super assistant.
>> Is this a is this a technical term by the way?
>> No. Yes, this is absolutely technical term. Sometime we not to be confused with the super duper assistants and those are we see those as the frontier clearly.
>> That's AGI.
>> Yeah, that's hi. Exactly. And so where these agents are really um representative of is they're almost like an extension of my identity. A lot of them the agent runtime runs on my computer. It has access to all of my same tools. It has access to all of the different things and applications that that I leverage. These things are magical, right? Because I can suddenly ask an agent, hey, look at my calendar.
Where are their opportunities to streamline my day? I can go in and say, "Hey, go in and reach out to these different members of my team that are managing a project and consolidate all the updates into a summary for me." I can go in and say, "Hey, I need to go and schedule a meeting with this person.
Go in and do the coordination back and forth in order to arrive on a time." I think there's going to be a a period of time where in the future we're going to think it's crazy that we used to email back and forth multiple turns to try and find 30 minute slots where we could speak. But that's one agent category and it's booming and there's a lot of really interesting developments that are going on. Now there's a lot of security and governance challenges that people are also saying, you know, like how do I think about identity in this world where it's not clear if it's the super assistant acting on my behalf or if it's me directly. and there's a a lot of questions and a lot of really interesting um decisions that are going to come about. The other category that I'd say is really more of these sort of workspace agents. And you could almost say the metaphor there is the AI co-worker. And it's it's kind of like, hey, we have this process in legal where there's someone who's hired that goes in and every time we review a new agreement, they have to go in and see if there's any non-standard terms and they go through the process of pulling those all together and then do a review of them. And there's a whole triage process that happens as a result of it. those types of things can be very much automated now with the new capabilities of these agents um with the new capabilities of these harnesses. Now, it doesn't mean that we're still not going to have people that are human in the loop to go in and judge exceptions, but for a lot of these processes where maybe 75% of it was just evaluating are the terms standard or are they non-standard, that should just be offloaded to the agent. and we're going to start working with those more and more um as co-workers. And that means that they might not just be locked in these chat interfaces, but they're headless and it's just another person on Slack that's got a funny emoji and responds a lot faster than some of my other co-workers.
But I think that that's where we're starting to see these agents now as they're capable of more and more knowledge work really enter the workforce. Obviously, you know, the reason we're incredibly excited by this is enterprises are sitting on about 90% of their data as unstructured data and this is all of the enterprise content that has, you know, research information and obviously you mentioned contracts and financial documents and HR records and policies and road maps and so getting the agent access to that content in a secure way in a way that the human still remains in the loop to be able to do that final contract review. How should an enterprise be working through this problem right now? What do you guys advise your customers?
>> Yeah, I mean I think that there's a couple different levels of context and what we can mean by context. I think the first and most obvious is just do you have your data available to these agents and that's the connectors and hopefully companies like OpenAI are doing a really great job so that that's really easy that there's a whole proliferation of different connectors that cover everything from your data systems to your line of business systems to your unstructured data. There's a whole category of customers and and a huge portion that leverage content management tools to be able to go in and make sure that that information is available and box is a great example of that. And so I think number one is just making sure that you have the data access and and that's something that whether it's MCP or whether it's through internal APIs or even now sometimes you're seeing CLIs via a method to be able to do that.
That's that's really important. I think the second piece though that companies are struggling with is okay I have this data available but what are the underlying processes that actually drive the creation of the data into these systems or these unstructured data and I do think that that's an area that I would argue is still emerging I've talked to companies that are trying to figure out how do I extract core business context in this in the most seamless way and They had even talked about doing long- form interviews with all of their workforce that's reaching retirement age and like like we don't know there's a ton of institutional context that's really important to the business and we don't even know what's valuable but let's at least try and extract as much of it into a structured format as possible by doing long form interviews. And then there's companies that are interested in just do we do business process monitoring or do we have agents themselves start going in and learning. And I think that both of those are going to be necessary to be solved. You're going to need to have access to the right data systems. But then you're going to also need the business context because you don't want an agent saying when they get the task come back and give me the previous six months of revenue. Well, that company might calculate revenue in a certain way. If they say, "I want the last six months of revenue in Asia-Pacific." The way that that company defines Asia-Pacific might be different than another organization. And so without the business context, it actually makes it very difficult for those agents to do the highquality types of tasks that ultimately end up being able to then drive the automation that is people have the trust and the the feeling that this is truly like an AI co-orker that deeply understands this business, not just connections to the data systems. When you think about the future of software and especially this idea of headless software because you know we we we see a world where a customer could either come to box directly and use the box agent directly in the box environment with something like you know GPD 5.5 and be able to leverage all the the uh the great capabilities within that the model directly within the box environment. And we equally see a world where you could go to chatbt and connect into box or go to codecs and connect into box and be able to access either the box APIs directly or even our agent directly and that's feeding in answers and information into a chatbt system. So we want kind of to make sure our customers are clear that that you know complete interoperability is is super important for us. How do you think about the future of headless software? um what are the kinds of things that that enterprises should be thinking about when they think about these these sort of now new API first ways of of working with technology?
>> Yeah, I trul truly believe that headless is going to be the way that the market's going to go just because in one sense the agents now are so good at software engineering that they can build the front end. They can build a lot of the way that you might want to represent the data or the type of task or the type of action that they need to be able to operate on. And the reality is you want that agent to be persistent with you wherever you go. And so it' be really weird to say like, "Hey, there's this amazing co-orker, but they can are only available at this period of time during the hours." And uh they only go in and log in through this web application.
That's the only way that you can go in and interact with them. The reality is like there's co-workers that I'm going to interact with via Slack. Sometimes I need to text someone. are sometimes that I need to go in and give them context and I want to be able to ask their opinion while they're over my shoulder and we're both looking at the dashboard together and I think that the reality is that agents are going to be similar where they should be able to be persistent travel with us wherever we work and that's going to be a variety of applications that's going to be a variety of different communication types. One final thing and you kind of brought it up around you know access controls and should the agent be your identity and kind of access everything that you have access to or should it be se separated as its own separate kind of identity. Obviously within the box environment things like you know file permissions and access controls and data governance becomes very important because you want to make sure that you know what the agent did with your data.
>> You want to ensure that you haven't kind of over over permissioned your environment. How should enterprises be thinking about data governance in the world of agents? I think that right now we are at a stage where companies are being less risk averse than they will in the future in the sense that you have a lot of bottomup experimentation and companies want to be able to encourage that because there's so much value in getting your workforce to be able to understand deeply this technology and be able to understand how to apply it to their work. But I think that there are pieces of that security stack right now that have not solidified at the same rate that the agents are improving. And so to your point, you could actually impact this into a lot of different ways. So first, there's the fact that agents are running on the computer itself and able to make changes to those core file systems within the the computer. Then there's the ability for the agents to then go in and perform tasks on behalf of the user. And so you want to have the custody chain of auditing across like was it the user that directed it? Was it an agent that went in and did it? And how so? And then you want to make sure that user permissions aren't superseded. Now that's good when it's a personal agent because it's effectively inheriting my permissions. There's going to be a question though of like does it also inherit my actions? Am I responsible for the actions? A third area where I think there doesn't seem to be yet a solve as well is this idea of the this democratization of software engineering.
So what we're seeing is that one of the killer uses is you can use this agent to be able to pull a bunch of company data, build a just in time dashboard that I can use in order to be able to make a decision or order to articulate a perfect a part of my strategy. And there's this suddenly what do you do in this world where suddenly everyone's able to create software and now it's like maybe the agent actually followed the permission set. it had access to data that I had, but then I published it to a dashboard that I've inadvertently made available to a bunch of people that don't shouldn't have access to that data. And so I think that we're hyperfocused on this problem because there's going to be a period where people are going to move from this experimentation to I actually want to build a strategy for deploying this at scale and I'm going to need to have chain of custody around the identity.
I'm going to need to think about how do I secure all this proliferation of applications. I need to think about that business context layer that we're talking about. Where's that codified and the permissions around it? And so we've been spending a lot of time building a lot of features and a lot of infrastructure just to be able to solve those problems because even things like skills more and more when I talk to companies that are saying how do I make sure that I just even have a governance model around all the organic experimentation you know I've got 30 different skills within my finance team like who's actually being able to go in and assess which of these are the most valuable and there's a discovery aspect to that but there's also security aspect of that.
>> Yeah, I mean skills will eventually contain uh proprietary information that can't be broadly shared. So yeah >> and then you add memory to that as well and and and I think things get really complicated quickly. So what I would say is there's no silver bullet at this stage in the market. We as a lab are maniacally focused on solving a lot of these problems because it's going to be an inhibitor to scale and adoption and there's a massive opportunity to provide that in the interim the the short answer that I would say is that being able to make sure that ident that these agents are inheriting the identity that you have good structure around all of your core data systems those like kind of the foundational basics everything that you would apply to your employee population applies to this agent population. And then I do think that the good news is that the cavalry is coming and that we're thinking a lot about these problems and there's going to be some really interesting ways that we can help with skill governance with with agent memory with proliferation and uh the management of company context and that we're thinking about that as a a key part of solving the problem.
>> Yeah. I mean, what I uh I I think I also heard in there and at least um generously interpreted is is that you really will want to have a secured and governed file system because a lot of that knowledge has to show up at the right time in the right way in a secure way for uh for your agents. All right, Dom, thank you so much for joining us.
Appreciate your perspectives and uh we'll see you soon.
>> Sounds good.
>> Thank you.
Every organization's knowledge lives in their collection of contracts, documents, charts, videos, and every other type of unstructured data.
Companies need to extract that important business information so that intelligence can become action. The work our teams do produces huge amounts of missionritical information that lies trapped in unstructured data. Accessing that information efficiently is the key to unlocking productivity, but it lies in millions of files and they're all of different formats. Your legal team might be spending significant time finding and tracking the critical terms and conditions for thousands of commercial lease agreements that are each hundreds of pages of legal ease. Teams handling routine shipping must read scanned handwritten notes, stamps, or images to confirm shipments. Each one is then manually updated into a system of record. Let's think about lending. The faster teams can extract information and make quality decisions, the more loans they can make and handle. Today, many of these contentbased processes are manual because existing tools are complex, brittle, and expensive to maintain.
These existing tools are mostly point solutions that create content silos and security and governance headaches.
Worse, they simultaneously swell tech stacks and leave many processes unsupported. And that's what we set out to solve with Box Extract, a powerful new agentic data extraction solution designed to extract actionable structured data at scale from your most complex content. Box Extract combines the latest AI models from Google Anthropic and OpenAI. Advanced OCR capabilities and agentic approaches that understand document structure and meeting to automatically and accurately extract information from a variety of content and save it as metadata alongside the content in Box. Process owners can configure, customize, deploy, and manage data extraction processes via a dedicated UI and leverage populated metadata to automate business process workflows. Accelerate content discovery and enable teams to make smarter, faster business decisions. For this to work in real business process, you must be able to trust the data. So how do you go from AI generated outputs to data you can act on? We think about this in four layers.
Visibility, confidence, control, and scale.
Bounding boxes in our UI give you a direct line of sight from extracted data back to the source in the document.
Instead of guessing whether something is correct, you can instantly verify it, building trust and speeding up review.
But seeing the source is just the first step. With confidence levels and human in the loop review, not only are we unlocking the ability to measure accuracy, but we're also enabling teams to drive action and determine what we can move forward automatically and what needs further review. When confidence is lower, those fields are automatically flagged, enabling users to instantly review, validate, or correct them without slowing down the entire process.
So you can scale automation without sacrificing control. And once you can trust the data, the next step is scaling it. With box extract, AI doesn't just extract the data, it helps you define how that data should be structured. Soon we will be introducing AI generated metadata templates, unlocking the ability to automatically generate metadata templates, including fields and extraction instructions directly within Box Extract. So you can go from unstructured documents to standardized reusable metadata models in seconds. No manual setup in the admin console, no context switching, just intelligent, structured, ready to scale across your enterprise and other Box workflow automation products. Let's see Box Extract in action. Box Extract lets you capture your mission critical data in a matter of minutes and store that rich metadata alongside your content in Box.
We begin by creating our own custom extract agent and giving it a name.
Users can leverage existing metadata templates or create new templates that map their structured data to specific metadata fields. Soon, you'll be able to leverage the power of Box AI to automatically generate templates within Box Extract by selecting several reference documents.
Once the agent has created the fields, you can navigate through each individual metadata field to view or modify those field properties.
Now that the metadata template has been created, users can then select their extract agent of choice and determine which source folder to apply this custom extract agent to.
Any contract that exists within that folder or that is uploaded into that folder will automatically have structured data extracted from it and saved as metadata alongside the contract in box.
Now let's add some contracts to the completed contracts folder. We can cycle through each contract, view extracted metadata, address any fields that were flagged as low confidence by our custom extract agent, and choose to accept, delete, or modify the extracted values accordingly.
And that is how you unlock critical actionable data within your contracts faster than ever with box extract.
That was a powerful example of AI extracting data from unstructured content. The critical first step in transforming your business processes with agentic workflows. Hello, I am Nurmal. I lead product management at Box and I'm here to show you the power of adding agents to your business processes. Using agents for a single task is very powerful. But the real breakthrough actually happens when they are natively built into your business workflows. So work accelerates and doesn't pause every time a decision is needed. Every enterprise runs on hundreds of workflows and many of those like loan applications or vendor onboarding are heavily dependent on people at critical decision points.
Let's take invoice processing as an example. What you see here is a very basic but typical workflow.
content comes in, someone reviews it, routes it to the right people for approval and at the end someone does the final validation. So what slows down these processes? Let's take the first decision point. Invoices usually come in different formats, PDFs, emails and scans. And right away the process slows down. Someone has to open the file, read it and extract key data. And that might take about 10 to 20 minutes of effort.
But the bigger slowdown comes from the time before work even begins. The invoice sits in a queue waiting for someone to have the context and the capacity to act on. And that's your first bottleneck. The next one is routing where someone needs to review the document, decide where it goes, which approvals are needed and that adds more waiting to the flow. Then comes validation and final processing. Again, another step where the workflows can stall. So, one invoice could take 30 to 60 minutes of effort, but it's wrapped in hours or days of waiting where every handoff slows things down and introduces risks. And when you scale that across the business, those roadblocks start to add up, slowing processes down and leading to lost work through manual effort, delays, and errors. And Box Automate absolutely changes that. Box Automate is our agentic workflow automation platform designed to remove bottlenecks in critical processes so things move faster and more work gets done. What's different? Box Automate is natively agentic. Most automation platforms treat AI as a step you add into your workflow. For Box Automate, AI agents are first class participants in the workflow itself. So the processes doesn't just execute tasks. It adapts and move work forward in context.
Automate orchestrates content workflows across your business critical systems combining agents, workflow logic and human reviews. AI agents are not just helping individual tasks but can be built, customized, deployed at scale across multiple workflows. And because it's built on Box, it connects natively with forms, do genen, apps, and sign while also extending seamlessly across third party tools. So your workflows run across your entire tech stack. Here is what that looks like in action. As invoices comes in, Box Extract captures key details, vendor, amount, dates, all in seconds. And from there, the agent keeps the workflow moving. They validate completeness, apply business logic, and even determine routing based on content.
Agents can also support summarizing documents, flagging anomalies, and even preparing your outputs for human review.
What used to get held up in decisioning now happens in minutes with far less manual coordination. And this is where the force multiplier kicks in. At the task level, delays in movement could go from days to minutes. And when you scale that across departments, let's say you have a 100 invoices per month sped up by a day, that's about 100 days of work returned to your team every year to focus on high impact activities.
And speed alone creates risk at scale for your business. Human in the loop review ensures that every outcome is consistent, accurate, and trusted. And with automate, as you combine agents with human in the loop review, you get the speed and efficiency you want and the control and reliability you need.
And this is exactly how Box Automate transforms your workflows and drives scalable automation. What was effort inensive becomes autonomous. What was time consuming becomes accelerated and what was errorprone now becomes accurate.
In short, automate gives you the intelligent workflows that moves faster, operates more efficiently, and delivers more consistent outcomes.
Now that we all know the value of Box Automate, let's see it in action.
>> Imagine you're a lending operations manager. Every day, your team reviews hundreds of loan applications, plus supporting documents like payubs and financial records. It's a very slow, time-consuming, and manual process. What if Box Automate could handle the intake and analysis while keeping your teams in control? The moment a new loan package is submitted to Box, the workflow kicks off automatically.
A risk assessment agent reviews the application and supporting documents, extracts key financial data, and evaluates the loan against internal risk guidelines, and in a matter of seconds, it returns a risk recommendation. Applications with medium or high risk are routed to a loan or risk officer, while lowrisk applications can move forward automatically.
The reviewer sees everything in one place and can instantly approve or override the recommendation.
So, in just a matter of a few steps, Box Automate turns a manual loan review into a faster human verified lending decision.
All of these workflows have resulted in hundreds of SAS applications and custom enterprise specific implementations. A recent study shared that large enterprises average over 690 apps. And customers don't want this spend risk or complexity. Instead of buying custom applications by departments or teams, they want to be able to consolidate on platforms with the flexibility to serve multiple use cases. And this is even more critical now given the need to have your content and therefore your context available for the right agents when they need it. All of these content applications need the same building blocks. They need storage, governance, workflows, integrations, intelligence, and a flexible UI.
Box gives you all this foundation, and Box apps lets you build these apps instantly. Box apps is the flexible application builder and space for teams to run their processes their way. It lets teams pick and choose from all the capabilities we've shared today.
Box apps is already live and we will enable users to build their apps by talking to a Box agent. This will make conception to deployment as seamless as possible as teams automate more of their work processes.
With that, let's now take a look at the app builder we are developing.
Box Apps puts purpose-built tools right where your content lives. Need something new? Just describe it. Here we're asking Box Agent to build a contract management dashboard. No templates, no developers, no waiting. All powered by the Box platform with a scale and security guarantee. In a matter of seconds, you'll have a complete working application. Visualize that data that's in your documents, monitor your business workflows, and run them all with an easytouse interface. Customize it to fit your needs. like filtering, searching, and sorting. Just say the word and it rebuilds on the fly. Every component stays connected to your actual box content. Go beyond dashboards. Set up actions and trigger automations in the same conversation. A weekly notification trigger done. No switching tools, no code. Describe it, refine it, automate it. Your next business app is one conversation away.
Customers are already seeing tremendous value from Box apps and e advanced capabilities. For example, a professional services firm sped up their client onboarding by about 40% using Box Extract and Box apps to automate their MSA generation review and then managing that through secure external collaboration workflows. A CPG consumer goods customer had scattered creative files with inconsistent naming and missing metadata. This created costly errors in managing rights and expiration windows. Using Box Extract and Box apps to autotag images, they saved significant time in tagging and finding assets. Reduced costly errors by over $250,000.
The contract customer is now deploying procurement automation and the CPG company is adding compliant workflows.
To explore this further, let's hear directly from a box customer. I'm super honored to introduce Evelyn from Samsung to share how Samsung is bringing AI powered workflows to their enterprise.
>> So hello everyone. Thanks for having me here. I'm Evelyn Yay. I'm the head of the GRC government risk and compliance as Samsung semiconductor.
As for Samsung semiconductors over the past 30 years, we are the industry leaders for the memory product. This is being utilized by the smartphone and the data centers and we are for the US headquarter. We are located in California, San Jose and we have been not just in the US we have international location as well especially for the R&D team. We have locations in South Korea, in China, UK and India as well. And for us, we try to enable the innovations growth across the high-scale data centers, automative as well as mobile and others electronic supply. We want to enables them especially around the AI infrastructures for using our memory chips as well. When I first start working at the Samsung couple things I see the challenge is they have a lot of the manual process we have to overcome.
Couple example is as a head of GRC we have to deal with a lot of the supplier.
We I work with the supplier to making sure that they are on the secure level is as meets our Samsung semiconductor requirement. We want to make sure that they are secure enough when we are using the products. So another thing that I also see within the Samsung that I having seen difficulty is the employee data. I see a lot of employee data saving in many different occasions. They could be on the network drive. They could be even saving saving on the people laptops. That's the part that concern me is because of how sensitive those data are that could be like you know we want to make sure that it matchs the regulations. And when we talk about the third party risk management the way that we manage why I call it that's a loss of many processes because we have to look at the different type of answer that suppliers responds based on my questionnaire and then we calculate the score with the score we're looking at whether those vendors are critical or high critical or is a medium based on the credalities we determine what question we are going to ask them what's information they have to provide to us all of them we go through just by mail.
So that is very like know time consuming for us by thinking about that. So we've been looking at many different solution we try to use. We look at different apps and also we try to even see whether we could home build any solution to help automate those process. At the end we're looking at the box which we have been using for a while but we find out that box have a lots of the new features that we could use to actually automate a lots of this type of process from end to end.
Everything is contained within the box for the information and we could automate everything within that using the box features. So the way we kind of looking at the third party risk management is if you could see as before we'll send out the informations by email or having people just like you know manually answer my email as a questionnaires and then now we got using the boss form to have the user to fill out the information and then behind the scene once they fill out the informations they could calculate the score for us. Now I don't have to even look at all the answer that they have been filling out and from the box AI agents they come up with a score for me and then from the score then we determine which vendors I have to send set up which type of questionnaire or request by using the file request from box everything I talk about right here we it's within the box app so I don't have to even using another third party application to support it as you could see because doing this type of automations by using box AI as well as box apps, people could fill out, you know, informations quickly. And then for us, we could review the informations by just looking at the metadata. That saved us a lot of time from originally it took took us like you know three days for my team to just look at one vendors and now it could only took us like you know half a days to do that. And for the employee data now I also using the AI they come up with the metadata and again to determines the the retentions based on the expiration dates we delete the data we don't necessarily need it and then that reducing a lots of the cost as well as making sure that we're complying with the regulation as well as you could see it back to the third party applications they fill out once the user fill out this is where the I talked about behind the scene the AI already putting all the information in and calculate the score for me. All I have to do is looking at the metadata from what's the AI agents already come up with and then determines where does those information are correct. We so my team is just validating what's the AI agents come up with and then with that from the box dashboard we pick up like you know which vendors we want to look at and then we will start looking as the vendor's informations and as well as the the metadata and the determines those vendors whether we approved to use those vendors or not. So what next as I mentioned with the box it help me reduce a lot of the times on already two use cases one is around the employee datas that is scattering around I could use that to kind of help me to do the data retention or data archiving and then the other part is the risk management for the vendors it's reduce my times for using the manual process from sending out email reviewing that every single answer manually now I could use the box AI and and then also the box apps to to streamlines the approval process filling out the informations as well as the calculating the risk score. So now we determines like you know there might there are others department really like to also leverage the box as well. So we've been working with the tax departments about on the a on the tax form so that they could have like you know they don't have to open up every single tax form right now. They could use the AI the the box AI to reviews to analyze the datas for the past few years of the tax incomes. Now they streamline a lot of their process. They have like you know multiple years of the document.
Now all they have to do is just go into the box AI and then ask the questions and then they will be analyzing for them. Another part I want to to use that to protect our companies on the sensitive data is using shields pro. The shield pros have the shield pro classifications based on that we could set a lots of the access controls around that to making sure that the type of data who should be accessing those. And the last part box automates I heard this just came out. So we are going to use the box automate to help us a lot of my employee on boarding and offboarding process. So of those sometimes we're using Excel spreadsheets or or task to track like you know so once employee and boarding which like departments should look at which form and then now we're using the box automate they could automate the whole process and sending out the document to the right departments to do the onboarding and then when the the employee leave using the similar methodologies to offboarding the employee. We are excited to looking at more and more new features from docs so that Samsung can leverage them. Thank you.
>> As you can see, we are entering an era where successful enterprises will have an exponential increase of AI agents operating across every department. As we know, these workflows can span across organizations and multiple systems. Yet, each of those systems and their agents needs governed access to content and context. Every new application built with Box, every new integration benefits from Box's content platform. This is why we are the perfect partner for significant developer communities.
Enterprise developers need their content and agents integrated across their tech stack. We are also seeing more enterprise developers in individual departments and with core engineering developers. ISV developers need a readyto-go and fully scaled content layer and partner developers both large, regional or vertical focus need the tools to quickly tailor solutions to unique client requirements.
All these developers need a govern content layer for agents. Files are the native unit of work for agents. Agents use files as context and produce files as output. By building with Box, they instantly benefit from our 20 years of experience. This year, we are continuing to build a frictionless, compelling developer experience, one that's fast to start, safe to ship, and offers both sandboxes, and improved docs. We're also investing in building scalable capabilities like enterprise retrieval, content intelligence, and MCP integrations to deliver end-to-end workflows. Building on Box will be the simplest choice for developers to apply AI to content. Our developer platform is the key content intelligence engine for this agentic ecosystem. Box MCP and A2A acts as a single secure bridge connecting external AI agents directly to customers content in Box. Instead of building complex custom code, we're standardizing how thirdparty agents like Claude, GitHub, Copilot, and OpenAI access enterprise knowledge while strictly respecting existing Box security and permissions and ensuring you control the knowledge agents contribute to. We also have over 1,500 existing integrations. So, Box content and Box AI is already plugandplay across the entire enterprise tech stack. Last year, we launched critical bridges to the most powerful AI ecosystem players, ChatGpt, Google Enterprise, Claude Skills, and the Box AI agent for Copilot. And we're continuing to invest in integrations in fiscal year 27 with upcoming launches. Each connected platform enables more valuable use cases of the Box platform.
Let's watch what's possible with our MCP integration.
>> The Box MCP server is a secure bridge that connects AI agents to content stored in Box. Let's see how this works in action with the MCP connector with OpenAI's chat GBT. We start with simple request. search box for Solara Energy files and summarize the company's business model and strategic priorities.
Instead of manually hunting through folders and reading documents one by one, the AI agent used the Box MCP server to securely access the right files inside Box. Now I have a clear understanding of how Solara Energy operates, what they're focused on, and how they're performing all in one step.
Now I'm taking it a step further, extracting data from unstructured documents. Instead of just reading insights, I'm extracting them into a format that could be reused in dashboards, CRM systems, or reports. And finally, it turned all of this into a deliverable by drafting a one-page executive briefing with recommended next steps, saving that new document directly back into the same box folder. So, we've gone from discovering content to analyzing it and structuring it and now creating a polished output end to end, all without leaving your AI agent of choice. By connecting your AI agents directly to content stored in Box, the Box MCP server ensures content is always fresh, governance and permissions are enforced, and sensitive files never leave the Box environment.
>> Hey, I'm Vishi Ganesh. Uh, I lead the AI platform division at Paychecks. I've been with Paychecks for over 16 years in various capacities be from engineering to architecture and now I lead the um AI platform division. Just a bit about paychecks. Paychecks is a digitallydriven HR leader that is reimagining how companies address their needs for today's workforce. We are more than just payroll and we deliver HCM solutions at scale. We manage over 800,000 businesses and 19,000 employees and we pay one in 111 employees in the private sector. We've had more than 140,000 clients who've been with paychecks for over 10 years and we've been in the industry for over 50 years with highly experienced HR professional reps. So to go to the problem statement, paychecks had a pretty complex life cycle when it came to document management. The diagram today shows how even a single document moves across many different personas and different channels.
Customers, sales rep, and customer success teams each interact with these documents at different points in the process using different interfaces.
This creates a complex document journey that is difficult to govern, track, secure, and scale consistently. The diagram represents only one version of the problem. Across the organization, different teams have built similar document flows based on their own business needs. Each solution was optimized for a local process, not for document management as an enterprise platform.
As a result, we've had multiple stacks, duplicate patterns, and inconsistent user experience and a fragmented storage and collaboration models.
Document management is an important business capability, but it's not our core industry. Our goal is not to become a document management company. We needed a scalable, secure, enterprisegrade platform where document handling is a core competency of the provider while allowing us to embed that capability into our own products, workflows, authentication models, and client experiences.
So what we decided was we chose box after a number of competitive analysis and here are a few things that kind of clinched the ecosystem for us. Box provided a streamlined and standardized way to acquire documents. Clients and internal teams can now upload, receive, collaborate through a consistent model.
This reduced the need for teams to build its own document management stack. It also gave us a pattern that scaled across multiple business processes at the same time. It still supported the specific needs of each life cycle.
Box had to fit into our mature enterprise ecosystem. We couldn't afford to build an ecosystem around Box, but Box had to fit into our ecosystem.
Paycheck applications could continue to own their user experience and business workflows.
They were also able to continue using their own user management, authentication, and authorization flows.
By abstracting Box's APIs and events, we could bring Box's capabilities into our products. This let us use Box without forcing teams to design around Box's native interfaces.
So, let's talk a little bit about the governance layer. So, Box's governance helped paychecks manage retention, legal holds, disposition, and document life cycle policies.
Box Shield helped paychecks protect sensitive client and business documents through classification, threat detection, and access monitoring and policy based controls. This gave us a strong visibility into risky activity and sensitive documents at scale. Box collaboration helped client implementation teams, sales service, and internal stakeholders securely review and share documents together. These capabilities allowed paychecks to apply governance, security, and collaboration controls consistently through the document life cycle. So what was our outcome? So we integrated box as a platform capability within the paycheck ecosystem. The core platform handled integration of Box eventing, metadata, and user management into the ecosystem while still allowing consuming teams to use Box's native APIs where needed. Different consuming teams could then reuse the same foundations within their own life cycle. This reduced the need for each team to build a separate document management solution.
Internal users and external client users could collaborate on the same document with bespoke permission models for different user types. Box's metadata eventing and life cycle capabilities offered a very rich ecosystem and that helped us scale solutions more easily and extended across different use cases.
Box's integration will allow us to scale the platform across more than 200 plus engineering teams, 10 business units, and 800,000 clients.
So digging a little deeper into the box eventing ecosystem. So we wanted to see how do we convert boxes event into business events. So, Box provided a very rich event model with over a 100 event types across user events and enterprise events.
These events were important because they allowed paychecks to track document activity at scale and respond to business context. We structured the box implementation to align how paychecks wanted to scale document management as a platform capability.
The goal was not just to consume the raw box events. The goal was to translate those events into meaningful application and business events. Our event listener and orchestrator pattern captured box events and converted them into paycheck specific workflows. For example, collaboration events supported approval flows. Metadata event supported notification flows. Shield event supported security and risk flows. This allowed Paycheck to use Box's eventing model while keeping the business process orchestration and application behavior within Paycheck's ecosystem. In short, Box generated documents events.
Paychecks converted them into business actions.
Mapping Box's user management to Paycheck's identity was very critical.
This slide shows how we used Box's user model while keeping Paycheck's user identity as the source of truth. We wanted document management system to be transparent to our end consumer.
Users continue to work within the page experience while Box supported the document management behind the scenes.
Each document upload, review, approval or a collaboration event had to map back to the actual user and not to any anonymous activity.
We wanted transparent transfer of user identities across the whole document life cycle. This allowed PEX to preserve its existing identity model while using Box's document platform. It also gave us clear auditing across both layers that is the page user identity and the box document activity.
So what are our next steps? So from an AI platform perspective, we want to integrate page's agent with boxes agents as part of our agent architecture. The goal is to seamlessly collaborate between internal page agents and third party agents. This includes orchestration across Box agents, Salesforce agents, and paychecks business agents. From a Boxhubs perspective, we want to integrate with Boxhubs to organize and extract metadata at scale. This can support ingestion of client documents such as tax documents, reports, and time sheets. The goal is to help clients get more value from the documents they already utilize. Today box hubs can help surface insights from the client information in a more structured way from a internal productivity perspective. We want to bring box AI concepts into the paychecks workflows for internal productivity.
This can support teams such as legal, finance, engineering, and operations.
The goal is to reduce manual effort and improve how teams search, review, summarize, and act on documents.
To kind of wrap it from an overall narrative perspective, the broader goal is to connect Paychex's AI journey with the best capabilities of our partners and what they bring to the market.
Partners like Box bring strong document intelligence and collaboration and life cycle capabilities. By blending these together, we can improve the client experience, enhance internal productivity, and overall elevate our users to spend time on higher value work.
Thank you and thanks for your time and hope you have a a good rest of the box event.
Hey everyone, my name is Mano Jastani and I'm the VP of product management for security, compliance, and governance products at Box. Content protection has always been a cornerstone of Box.
Autonomous agents create powerful new security risks and challenges. We're already seeing agent releases across the industry exploited overnight. In the last 30 days, three separate agent platforms were hijacked through the same attack vector prompt injection through content the agent was asked to read. In one case, an agent with database access silently exfiltrated data through a standard UI workflow. In another, there was no audit trail at all. When the forensics team tried to reconstruct what happened, there was nothing to find.
Traditional security tools are built for the network and identity layer, not for what an agent is being told to do inside a document.
Even agents that were deployed appropriately can be misaligned.
Enterprises will have to deploy agents against sensitive content with a trusted security and governance layer. AI has moved from helping humans with the individual tasks to automating entire workflows to now acting autonomously on your sensitive enterprise content. And as that capability grows, so does the risk surface. On this slide, you can see how our content lives across distinct silos and programs, each serving a different part of the business. External AI agents from thirdparty tools, partner integrations, and autonomous workflows are reaching into these silos to retrieve, process, and act on information. That's powerful, but it's also introduces real risk. This is exactly why security and governance must be foundational to the content layer rather than bolt it on as an afterthought. When AI agents are accessing your content, you need to know is this access authorized? Is my data protected? Am I in control? Box has a security layer architected specifically for the age of AI agents. And now we are going to give you even more control of how agents interact with your content.
Box is leading the charge on securing an agentic future. And we're excited to share plans for our comprehensive suite of security and governance controls purpose-built for the agentic era. We're creating an intelligent security and governance plane that lets enterprises safely orchestrate AI agents at scale.
Turning secure and governed content into trusted insights without compromising security for Box agents and third party agents that connect to Box. One policy set. No per vendor configuration required. Each of these capabilities in this suite addresses a specific failure point in how agents operate on your content today. Starting with input safety. Input safety scans every prompt and every document embedded instruction before it reaches the model. It detects known injection patterns and gives admins response options. Log it, alert on it, or block it. That matters because traditional security tools inspect what is in a file. But in the age of AI agents, you also have to inspect what the file is trying to get the agent to do. With action guardrails, organizations can set up deterministic rule-based controls that define the operational parameters for all agents, including agents external to box. No external sharing, no bulk deletions, no reading, restricted files. These guardrails are enforced at the content layer independent of the agents own logic. A single injected prompt cannot cascade into unauthorized actions because a guardrail exists regardless of what the agent is told to do. Action guardrails secure thirdparty agents such as claude chat gpt and copilot connecting to Box as well as Box agents being leveraged in workflows and custom agents in AI studio. Agent activity oversight gives customer admins visibility into what external AI agents are doing with Box content through visual monitoring and reporting. It tracks a volume of specified agent actions and helps teams quickly spot unusual or excessive behavior. Admins can define threshold-based alerts for selected actions, setting a specific volume limit within a chosen time window. When agent activity exceeds that threshold, Box surfaces an alert so security teams can investigate and respond quickly. This capability helps organizations move from passive monitoring to proactive oversight of agent behavior. By combining visual insights, reporting and configurable alerts, agent activity oversight makes it easier to detect abnormal activity patterns, strengthen governance, and reduce the risk of external agents taking excessive actions on sensitive content. Every agent action is logged with full session context, which agent, which user authorized it, which files for access, what policy governed it, and agent sessions are governed like content, legal hold, retention policies, disposition applied to agent sessions.
exactly as you would apply them to a document. If a regulator asks you to produce records of every AI agent interaction with a specified set of files during a specified time window, you can do it. Most solutions can tell that an agent accessed your content. Box is the only platform that knows what the agent touched, what was in it, how it was classified, what retention policy applies, and whether the action was in or out of policy. That is not a feature gap. That is an architectural strength delivering real security and governance for agents. Box is continuing to innovate to keep data protected with Box Shield Pro, our intelligent content security suite. This launched last December with automatic AI classification, threat analysis, and ransomware protection, which together help identify, avoid, and mitigate threats. This year, we plan on enhancing Shield Pro by adding capabilities such as label priority, a prompt helper, and retroactive classification. Retroactive classification is worth a moment here because it addresses a gap that many security teams are feeling. Classifying inactive content, the contract from 5 years ago, the HR records that have not moved in a decade, the legacy content sitting dormant within your organization. With retroactive classification, you can now go back and get visibility into the sensitivity of all your content, active or inactive.
Not just the files your teams are working on today, but every file in your environment. That means no more classification blind spots.
And for regulated industries, it means you can demonstrate audit readiness across your entire content corpus, not just a portion that happens to be in active use. Box Shield Protects your content with intelligent security. But for many organizations, especially those operating across borders, protection also means knowing exactly where your data lives. That is where Box zones comes in. Box supports data residency needs with secure and scalable in region storage through Box zones. On this slide, you can see all of our zones locations. We currently offer seven zones across nine locations covering Australia, the EU, the UK, France, Canada, Japan, and the USA. And we are expanding with Switzerland, Singapore, and Israel coming later this year. The value is immense. Your content is stored and processed in your region, helping you meet requirements like GDPR without disrupting how your teams collaborate.
Admins manage everything from a single console. User experience sees no change.
Data residency is no longer a trade-off between compliance and productivity.
With box zones, you get both. And that brings everything together. Intelligent security with Shield Pro, secure and governed AI agents, and now data residency with box zones. All of it built on the same platform, managed from the same console, protecting the same content. Deploy AI agents with confidence. Your content stays protected. Here's what that means for each of you. For the CISO, you get full visibility and control over every agent action. No custom pipelines. When your board asks what your AI agents are doing with your data, you have an answer. For the CIO, you have no more stalled AI pilots and you can move them to production. Security is no longer the blocker. One policy set governs every AI tool your organization uses. And for business owners, your teams can use AI to move faster because the guardrails are in place. Security is not saying no.
Security is saying here's how you do it safely. We're entering an era where every enterprise will have an exponential number of AI agents operating across every department, all working on your most sensitive content.
Each of those agents needs governed access. Box is the only platform built to provide that at the content layer for every agent out of the box. Box is the platform where enterprises can confidently adopt AI that is governed and secure by design. At Box, our mission is to power how the world works together. And we are giving enterprises the secure, integrated, and automated platform they need for their AI transformations. Thank you for joining us today to see how Box Powers, content plus AI, unlocking trusted insights, intelligent agents, and secure AIdriven work. We look forward to what we'll build together.
All right, that is just a preview of many of the exciting things that we are delivering at Box to help you transform with the full power of AI agents and enterprise content securely. Now, we can't do this alone. Uh so, we have brought together a uh partner ecosystem to make sure that we can help transform organizations anywhere on their journey with the full power of AI. And one of our key partners is Amazon Web Services.
And I'm incredibly excited to introduce Rahul, the VP of data and AI GTM at AWS to share a bit more about how we are working together. You guys have some breaking news with OpenAI and this kind of rocked the industry because now for as far as I can tell the first time ever customers can choose a different cloud to be able to use the AI models on through bedrock. So would love for you to maybe share a little bit about that partnership. We earlier today had Dom Gillo from OpenAI on stage uh talking about how we're working with OpenAI. So now there's a a clear kind of joint story as well there, but would love for you to share a little bit about what you're doing with OpenAI.
>> Yeah, thanks Aaron and it's great to be here. Appreciate the opportunity and um we are super excited about what we're doing with OpenAI. As you mentioned, it's a recently announced partnership and we're delighted to be able to bring uh OpenAI models um and so you can get the GPT models from AWS through Bedrock.
We're also uh working with them on managed agents and so this allows you to easily build stateful agentic applications uh that are powered by openai on bedrock and then open AAI frontier will also be available on bedrock as that becomes more broadly available. So, you know, as we work backwards from what customers need, obviously that choice of provider available on Bedrock alongside um Anthropics, a big deal, as well as all of the other model choices that we offer our customers and uh we're excited to partner with OpenAI and get this in front of as many customers as possible.
I think it'll be a big deal.
>> Yeah. Awesome. So, uh for anybody who somehow has been maybe off the grid for like the past month, very big deal. So I can as a developer or an IT organization right now or or imminently I'll be able to say okay I want GPT 5.5 running on bedrock and then go against my uh my my kind of cloud consumption on that front.
>> Yes, exactly. So you'll be able to select that as a model. Uh the beauty of bedrock is you can maintain stable APIs but pick your choice of model and uh this will now be available to uh developers uh coming very soon. We announced it uh just a couple of weeks ago and it'll be available shortly.
>> Awesome. Great. Well, yeah, congrats on that. And so now with our joint partnership as well, we're going to make it as easy as possible to be able to leverage bedrock services, uh, whether it's OpenAI, Anthropic, or other models.
And, uh, we're doing a lot of great work, um, with, uh, with with you guys on that front, which certainly brings us to AI in the enterprise. You, you guys have a, um, a very unique kind of view into the into the landscape just given the scale of the workloads you see, the kinds of customers you serve kind of across every single industry. would love a little bit of maybe u you know state of the union where are we with AI at the moment um you know there's a lot of talk of you know kind of we're beyond the the chat phase of AI moving more into the agentic phase what does that look like from you know your vantage point and with uh working with customers >> so what I see across close to a thousand customer conversations at this stage is that we're definitely in the year of uh Agentic AI I'd say in 2026 we're seeing um active production deployments for complex endto-end business workflows and you I think that is going to continue to accelerate. And you know, one of the key things that we see in the enterprise is that, you know, the key differentiator between customers that make it into production versus those that stay at the PC phase is the integration of enterprise data with AI. And that's really the only key differentiator that customers have. Anyone can access state-of-the-art AI models. It's the combination of enterprise data about the business, about customers, all of that embedded knowledge coupled with state-of-the-art AI that allows customers to build unique differentiated experiences. And we see folks uh, you know, saving tens of thousands of hours a year in clinical research from things like know your customer processes in banking and finance to claims, auto approvals, and insurance. It's really widely spread and accelerating at this point. You know, I think we've seen so much adoption and so much uh ascendancy on uh AI for coding and that that kind of is what we have a lot of the the best proof points for. As you think about we're moving from coding agents um to now more areas of knowledge work and and workflows. Where are you seeing the the the hit rate be highest? What what uh what do you kind of see as the as the areas that customers should be thinking about and focused on on that front?
>> Absolutely. So coding you're absolutely right that's you know taken off it's growing exponentially still I think we've seen a lot of early adoption in things like intelligent document processing in uh areas like support >> but I would say at this stage it's expanding to really every business workflow you can think of and so you know as we think about it from the AWS perspective um you can really accelerate any business process with agent assistance and so we've seen things as diverse as individuals prepping for a bi-weekly review with their CEO using agents to help them prep that material.
Today I'm using tools that allow me to generate a custom customerspecific presentation that incorporates all the latest information from CRM, all the latest news, all the latest releases.
And so I think we're really just limited by imagination, access to data and the APIs at this point. When you think about the um you know the areas that customers should be thinking most about where to get the most value from agents you know I think there's always this tension which is do I use AI to take a current process and maybe you know streamline it make it more efficient possibly reduce costs do I use AI to accelerate something where you know more output in that area could generate more value for my organization do you have any any kind of sense of where where customers or enterprises should be leaning where where you've seen the biggest kind of you know proof proof points for agentic success.
>> The way I think about it, Erin, it's it's really about um working backwards from key business objectives. So, I would say more about what is your business goal? Use that to then define what's the minimal set of data assets I need to light up that particular goal.
And then the third piece is um this idea that discipline equals freedom. If you set up good guard rails from a security, responsible AI and financial perspective, you can actually let the organization free to innovate. And so once you kind of have those three things set up, then it's about sprinting and really getting after it. You know, if you are early in your AI journey as an enterprise coding, aentic assistance for employees, customer support, and customer relationship management, these are great places to start. Uh I also think AI to eliminate drudgery is a great way to get started. So, you know, we used it to upgrade our Java SDKs and code that no one likes touching. AI was great at that. Uh, but if you're further along, I would look for key business objectives that you're trying to drive.
Maybe you're trying to drive down the cost of originating a loan. Maybe you wanted to find a new line of business or up throughput and then work back from there.
>> I got out of a meeting very recently where we finally um are updating a feature in Box that I don't think a single engineer has touched in 15 years.
And it's not because it's sort of an unimportant feature. It's just like the tribal knowledge, you know, sort of like eventually evaporated and nobody really kind of like knew how to maintain it and so it's been on just life support and finally AI >> is now the reason we can just keep innovating in these areas. Um and so whether it's you know the the migration project you never would have done before, it's the upgrade of an SDK, it's the new features that you couldn't have gone and triaged, I think it's this amazing relief valve in our road map and and in our normally constrained processes which is uh incredibly exciting. And so the thing that that we're excited about and I think you guys have have been on the forefront on and I've heard Matt uh Garmin and others talk about this is you know start to think about AI is really you know how how we have this new form of abundant you know intelligence. How do you use it to do all the things that you never could have done before >> as opposed to just taking today's processes and replicating them. So you guys are are right at the forefront of that and um uh and it's been awesome to see. Now there's there's sort of no free lunch in this world right now. And I don't even mean the cost of the tokens, but the the work it takes to make AI useful and and be able to get your organization into into an AI ready way um and format. How do you kind of see the upgrading of of infrastructure, data platforms, how companies need to think about their information, their systems, so that way we're ready for agents? It's a great great point and great call out.
I do think customers that really want to embrace AI are really thinking about how to modernize their data platforms and get themselves in order. You know, and often that means moving out of legacy systems and thinking about moving into more scalable architectures in the cloud. You know, often enterprises go, "Wait, I can't do anything. I've got to get my data house in order before I can get started on AI." And I think that's actually a mistake. I think you have to innovate while you modernize. And so you can do things like set up MCP access to data in its legacy form, start to use that to build your learning on AI even as you modernize simultaneously and then you can just repoint under the covers.
So I think one of the things that I spend a lot of time talking to customers about is yes, you do need to get your data sorted, but you don't, you know, don't wait for clean data. A, there's no such thing, >> and then B, today's technology with agents and models allows us to make sense of messy data in ways that we couldn't before. So figure out how to do both uh together.
>> We're seeing a lot of customers, I'd love your take on this, where they're ready to deploy agents. There's, you know, huge use cases right away out of the gate that will generate them more revenue and drive productivity up. Uh, but data is still stuck in in some of these legacy systems on in on premises environments. They're not in the cloud.
you know, maybe they're in platforms that are in the cloud, but they're not really built for agents. And are you seeing an acceleration to getting data ready in that kind of way? Because that's obviously an area that we're working together on uh that we're we're pretty excited by.
>> Well, you know, I I do think partnering together um is a great way for us to unlock customer data together. And I think, you know, the role that you play in sort of managing enterprise data coupled with AI and infrastructure that we can bring to bear is a great unlock for customers and represents a big opportunity. So that modernization acceleration is real and um one of the reasons we're super excited about our partnership. Uh and then I do think there are architectural patterns now with models and MCP that allow you to connect to remote data have that data be lit up for AI even while the modernization proceeds in parallel. So I think it's that you know the combination of the partnership uh unlocking of data that we have together that's in box but also finding ways to light up data that's not yet there. um even as it moves.
>> How do you feel about the level of of standards protocols in this industry?
You you brought up MCP. I think that's you know fortunately kind of we've converged on MCP. There's obviously a wave of of CLI based integrations as well. Do you think we're at a point where customers can feel that we've got a stable environment to to build on? Do you see changes coming down the in the horizon that we need to be prepared for on on how to you know build these systems? I think the change rate will continue, Aaron, in terms of models and specific flavors of agents, but I do think some of these interfaces are going to be pretty stable like MCP. I think we've all um committed to it. AWS sits on the committee and we're all working to continue to evolve that. And there are other standards that we also support, A2A, there's client plug-in protocols. So, I think this is a place where, you know, customers can trust the foundations that are around today. And then I think we will continue to adapt together as new things uh take hold and um you know really committed to kind of staying compatible with open source protocols in the space uh just to make sure customers have all the optionality that they need.
>> What are you seeing from the model provider and partner side? We've obviously had I'd say the past you know 12 to 18 months have been really kind of underscored by these reasoning models that that can you know obviously do much longer tasks. They're they're tied to a a harness in many cases that let them run for 30 minutes or hours in some cases. I have some agents that go off for for hours and come back, you know, with some some work product. How do you feel about model progress? Anything that that you sort of sense in the next six or 12 months that that we should be preparing for in terms of our systems and enterprise environments?
>> I think models will continue to advance.
I mean, we're seeing the state-of-the-art continue. the providers are all working to expand capabilities and so I think the one of the most important things we can do for our mutual customers is to help them design systems that can take advantage of new model capabilities as they become available. So that's that's one piece.
Another thing I think is really important for us to be thoughtful about is, you know, I'm starting to see customers not just automate a process, which I think gets you a linear gain, but if you can automate a process and then use the results of that to feed back into the process to make the process itself better autonomously, I think that you can get a compounding gain. And there I think that's a real opportunity. So now when I'm in customer conversations, I'm always trying to figure out how can we have this system not just automate but also autonomously improve itself. And I think if we are thoughtful about that, we can really build systems that get smarter as they get used more. And that represents um I think a big collective opportunity for all of us to go after. Yeah, I mean I think that will be certainly one of the big themes in the coming years is is the sort of self-improving feedback loop systems both at the model themselves getting better at at self-improvement but uh but can you design a an agentic environment uh in your workflow where you know once that agent performs its tasks is it is it writing back new best practices to some some form of a you know a skills file or whatnot? How do you start to then make sure your agents have access to that data? How are they how where where can they put that back in in terms of that feedback loop? Any interesting best practices you're seeing on how to design these uh improving systems?
>> You know, I think the way you put it is exactly right and I sort of think about it as it's almost like the reinforcement learning human feedback process and training applied to business workflows.
You know, I think it'll be about setting up the right guard rails. What's the right way to in a trusted fashion introduce new ideas into the system? And then how do you think about setting up your evaluation context so that you can be confident about what you're putting in is not going to take you backwards.
>> And so I think being thoughtful about designing that loop and then having an exception process to look at things that don't fit that form factor is going to be the way to do it. And I would say this is, you know, this is emerging, but I'd be surprised if we're not all doing this kind of at scale a year from now because it seems like a very logical extension of the the current path, which is get me from A to B faster. But this way we can start to reinvent the process itself, which sort of goes back to your point at the beginning about how to maximize the return that you get from your AI investment.
>> Yeah, I think that's right. And I think the big takeaway for all of us and and I think we're all learning this as as we go is, you know, we can all collectively raise the ambition level of what we can do with this technology. Um, and to do that, you do have to start to to think about redesigning the process for for a world of agents where where they don't automatically have the context where they do have to kind of be, you know, kind of put back on the on the rails if they if they, you know, go off and and they do need to have the the right human in the loop kind of elements to that.
But when you can redesign your process at the end of that, the gains, you know, can be enormous in terms of the throughput of whatever that workflow is and what you can deliver in your organization.
>> Yeah, totally agree with that. And um, you know, I think it's just an incredibly exciting time in technology because there's so much potential that's been unlocked. You know, I think the other thing that's super interesting and I think another reason I'm incredibly excited about our partnership is that, you know, agents have dramatic and AI has dramatically changed the amount of unstructured data in documents and files and images that you can extract and turn into signal that can be used alongside of what was classically lived in data systems.
>> And so there's that much more rich context available on which to kind of make better decisions. So that uh that I think was another massive unlock and huge part of the you know the value I think we can bring together.
>> Yeah. Awesome. Rahul, thank you so much for joining us today. We're very excited about the partnership. Um and to all the the joint customers out there, it's probably like a 97% overlap with uh with AWS customers within the box environment. We've got a lot of cool stuff cooking that we're happy to bring to you.
>> Love it. Aaron, it was great to be here.
Thanks so much and uh super excited to kind of go forward and build together for our customers. Awesome. Thank you.
>> Thank you.
Ähnliche Videos
BREAKING: Microsoft’s New Image Generating Model Beat Out GPT 1.5 and Nano Banana 2
aimmediahouse
122 views•2026-06-03
Long-Running Agents — Build an Agent That Never Forgets with Google ADK
suryakunju
142 views•2026-05-30
I Made the Same Anime Fight Scene in Every AI Video Generator
NobleGooseAnime
295 views•2026-05-30
Nvidia Bets Big On AI PCs | New Chip To Power Windows Laptops | Technology | AI Updates | N18S
cnnnews18
3K views•2026-06-01
3D Platformer Update - NO CAPES
SolarLune
294 views•2026-05-30
AI Doesn't Create Bias — It Inherits It
UXEvolved
176 views•2026-06-01
Distributed Inference Challenges Explained #shorts
alexa_griffith
466 views•2026-05-31
Starting & Test Driving JAKE'S Abandoned BUS from Subway Surfers | POV Restarting
RestartGaragePOV
4K views•2026-06-04
