Good afternoon. I am Cynthia Hiponia, Vice President of Box Investor Relations, and welcome to BoxWorks 2026 Investor Product Briefing. Since our comprehensive product strategy briefing that we provided at our Financial Analyst Day in March, our product leadership and pace of innovation has continued. I am sure you have seen our press release today regarding our exciting new product announcements.
Our executives will be providing an overview of today's exciting news and do a deeper dive on our Intelligent Content Management strategy. Then we will have a live Q&A session at the end of the presentation. Feel free to send questions to ir@box.com or enter them into the text chat box. During this presentation, we will not be providing any financial updates, nor will we be able to address any financial questions during the Q&A portion. Please note that this presentation may contain forward-looking statements that involve risks, uncertainties, and assumptions.
Further information on these risk factors that could affect our forward-looking statements we make in this presentation can be found in the documents that we file with the Securities and Exchange Commission. Finally, for a further deep dive on our product announcements, including product demos, customer testimonials, and a replay of this morning's onstage keynote with Aaron and Jensen Huang, founder and CEO of NVIDIA, I encourage you to head to our website. With that, I will turn the session over to Aaron Levie, Box's Co-Founder and CEO.
Thanks, Cynthia, and welcome to our virtual investor event at BoxWorks. I am Aaron Levie, CEO and Co-Founder of Box, and we wanted to summarize all of the major announcements that we have made today onstage at BoxWorks and really provide some extra color on how transformational these capabilities are going to be. At Box, our mission is to power how the world works together, and when we started the company, that was really about enabling people to work with each other and with all of their unstructured data. But now there is a new role that has entered this mix, which is the power of AI agents, and we are increasingly working with these AI agents to take full advantage of our data as organizations. We are incredibly proud because we get to work with over 60% of the Fortune 500.
We work with some of the leading brands and organizations all around the world, companies like Samsung Semiconductor that are transforming the chip industry, organizations like Morgan Stanley, which has transformed client workflows in their wealth management organization, the Mayo Clinic that is obviously delivering breakthrough research and clinical healthcare outcomes, and Toyota, which really has led the industry in advanced automotive manufacturing and innovation. What all of these companies have in common is they see the power of their unstructured data and how they can really work with that data to propel their businesses and organizations forward in the 21st century. What we are seeing across every single organization, every single industry right now, is that AI is fundamentally transforming everything about how we work.
Now, if you pay attention to any of the AI news and any of the AI progress that's happening, you've seen that we are on an incredible exponential curve around what we are now able to do with AI models. You just go back a few years ago to the ChatGPT moment, it was amazing because at the time, the technology already felt like an incredible breakthrough. But just in the past few years, we've seen this rapid acceleration in AI model progress. We got models that could reason, we got models that could execute multimodal workflows, we had models that could leverage tools. Now, with capabilities like computer use and much more, models are able to actually take on more and more advanced work within our organizations.
What we've seen is as the models train on larger datasets and with more compute power, we're getting greater degrees of capability and intelligence from these models. Again, just a few years ago, these were kind of useful assistants, and now we're seeing at the frontier, very soon these models will become superhuman in most domains of work, which is obviously incredible progress of the AI capability. Now, this is only going to be accelerating from here, because what we're seeing is that AI labs are actually approaching really a form of recursive self-improvement, which means that this is going to accelerate even faster. We ask ourselves, what would it mean if every single person or workflow in an organization had an assistant or a set of workers that could work a thousand times faster than we do?
What would it mean if they could review all of the data in a workflow, if they could look at all of our documents, if they could process transactions for us? How would that really begin to reshape work? We're entering this moment where AI begins to transform not just the underlying individual workflows that we have. Back in this early kind of 2023, 2024 period of AI where you might ask AI to summarize a document or ask a question about the internet or process a limited amount of data, but what if AI agents could actually process arbitrarily large amounts of data? What if they could read every document in a process or generate entire code bases worth of code for a project or look at all of the transactions in a system to detect fraud or any kind of anomalous incidents?
We're really entering this era where AI agents will be able to fan out and execute broad tasks for us. Now it's not just one agent performing one simple question and answer task, but actually swarms of agents or fleets of agents that go out and execute this work and actually go and accomplish goals for us. This means that the underlying workflows in our organizations are going to dramatically change.
This means that we are able to accelerate everything from a financial services project, where you could have AI agents go and read all of the due diligence on a transaction that is happening, to not just sort of sampling for due diligence documents and contracts and financial documents, but what if AI agents actually went off and read every single piece of data on that transaction and generated a packet of information to save months and months of manual work on that process? This fundamentally changes our financial services processes. What if you had AI agents that could look through all of the lab research that is being done, and all of the external research to an organization, and those agents can really, fleets of those agents are able to go out and do this work in parallel?
We will be able to get accelerated drug discovery within the life sciences industry. And what if we are able to dramatically do things like client onboarding or claims processing when you have the ability to throw compute at this problem where we can read through every document or look at every incident report to accelerate those workflows? So we are really moving from an era where we have conversations with AI systems to now AI systems are running in the background, executing on our work autonomously. And for the first time in history, we are now able to apply these agents to problems really across every single domain of work, and this is going to accelerate our work inside of all of these industries. What this means is that now knowledge work output is no longer constrained by the time that we put in as individuals.
We actually can begin to accelerate our workflows and drive greater execution and greater volume of output, really just based on the variable of how much compute we apply to the problem and our ability to obviously structure these workflows and enable people and agents to work together. So this has never really happened where we have had this discontinuous line between the amount of effort in and the amount of value out. We now have this ability to scale up output just based on the amount of compute that we throw in. We believe that in the 21st century, competitiveness for every enterprise will be defined by how well you apply superintelligence to your business, to your data, to your people, to your workflows. This is really going to be the new definition of how companies stay ahead.
The thing that we find is that models alone can not drive this transformation. Now, the models are necessary. It is critical. We want as much progress on AI models as possible, but the models themselves just are not able to deliver this, because if every enterprise implemented the same AI models, they would deliver roughly the same outcome. So there is an X factor, which is, how do we actually enable these models and these agents to work with our enterprise information, with our unique context, with our business decisions? How do we leverage AI in our organizations in a unique way? And what that comes down to is basically agents need context about your business. They need to understand everything that has happened in your organization so they can actually go and execute on this work effectively. This could be everything from our product plans and other information.
This could be the information in our CRM system or our ERP system. All of that is the enterprise data that we are housing as organizations. The really big job for everybody in IT right now is, how do we connect these AI agents to all of this data so they can act and execute properly? There is a really important piece to this, which is the Box piece. About 10% of this data is structured data. That is the stuff that lives inside of our ERP system. It is the stuff that lives inside of our CRM systems or databases. Insanely important data. But 90% of the information that we have as organizations is unstructured, and that is the data that we touch at Box and that we have been building the systems to help organizations handle. Let us just break apart that 90% for a second.
This is the data that has our product roadmap. This is the data that contains our financial projections. This is the data that contains our new marketing campaign or marketing strategy. It has our customer contractual obligations. It has our employee onboarding information. It has the due diligence information. All of that is the unstructured data that we work with every single day inside of organizations. We have this incredible situation, which is 10% of our data is structured, and it is kind of well-organized for AI systems, but 90% of this data, the stuff that really, really matters for AI agents, we have not yet really been able to have the architecture for enterprises at scale to be able to manage and work with that data, and that is what we are focused on at Box.
The problem is that a lot of this data is fragmented in a wide variety of silos in organizations. The data is stored in lots of different places. There are legacy network file shares. There are document management systems. There is on-prem infrastructure. Sometimes there are cloud systems. But you have a lot of fragmentation of this data, a lot of outdated and clunky technology. All of a sudden, enterprises want to go and deploy agents, and the problem is the agents are sort of going around and trying to find the information that they need, and soon we are going to have far more agents than we have people inside of our organization, and you are going to run into this massive existential challenge as an organization, which is, how do you get the right information, the right context to the right agents in a secure, governed, and accurate way?
Because what will happen is when agents are working on information that maybe is duplicated or there are different versions or they are in different silos, all of a sudden the agent might pull the wrong data. So you have a context problem, which is, as you have more fragmentation, the agent is liable to pull the wrong information or from the wrong source of truth because we have so much duplication and fragmentation. You have a much bigger issue, which is as you have more systems, it is very hard to keep them all in sync in terms of our access controls and our data governance and our permission models. We have a huge problem, which is all of a sudden, if you have a little bit of oversharing happening inside of one of those repositories, now an agent is able to leak information.
Whereas you have more fragmentation of these data systems, they are hard to protect, which means that it is easier for an agent to go and actually either intentionally or inadvertently leak data, or if there is some kind of attacker, they are able now find that information as well. Finally, you have a lot of interoperability problems. Sometimes these silos connect just to their own datasets in an effective way. We need systems that can work with all of our agents and all of our data in a secure and governed fashion. Enterprises need one platform that can connect their content to AI securely. This is incredibly important for their most important information, their most important content that powers and really is the underpinning of their most critical workflows as organizations.
That is what we are building at Box with the Intelligent Content Management platform that we have been advancing. It all starts with our global infrastructure, so we can power and help companies work with all of their underlying data in a fashion that works with their data's residency requirements that scales to any amount of information. We then have a layer of data protection and security capabilities, which really help you do everything from prevent threats and find anomalous activity and prevent ransomware events and be able to handle data classification or data governance. You are going to be seeing a ton of innovation at BoxWorks, and in just a few minutes around new advancements to help protect data in this agentic era. We then have a layer of content services that help customers power the full life cycle of their content.
The core file system capabilities, the workflow automation capabilities, our metadata platform, and you are going to see major advancements in our content services, both in the BoxWorks keynote and in a few minutes, as well as in our roadmap session that we did today. Finally, and really a lot of the crux of our major announcements is our agentic platform, and the layer that helps both prepare the data that enterprises are working with for AI agents, and is the AI agent harness directly built within Box. We do a lot of work on the first part to ensure that you can connect any of your enterprise content to any AI agent that is external to our system. This could be Claude or ChatGPT or Cognition or Factory or Copilot, Gemini, any AI agent external to Box.
We want to make sure that you can successfully and securely get the right data to those agents as efficiently as possible. We have a major new announcement around Box Context Retrieval that we will be sharing around how do we make that as effective as possible. We want to have the world's best agent harness directly in the Box platform for helping customers do advanced work on unstructured data operations. This could be things like metadata extraction. If you have 1 million or 10 million or even more documents that you want to extract metadata from, we have Box Extract, and our agent is highly effective at doing data extraction. That is one of the killer app use cases, obviously, in our agent platform.
We want to let customers be able to build custom agents directly within our agentic harness, or automate workflows, leveraging the Box Agent to be able to be deployed within our Box Automate offering as well. This is how you start to deploy fleets of agents in a workflow that can run as many times as you need, all governed with the right guardrails and protections around what those agents are able to do. Then ultimately, this all comes together with a major new announcement that we will be sharing around how you begin to design this full set of agentic workflows with our all-new capability, Box Architect, which is an autonomous system to help you build, deploy, run, and eventually optimize these agentic workflows directly in Box, leveraging our frontier intelligence breakthroughs from the industry.
We are going to share a bit more about these major announcements, and I am incredibly excited for you to see what we have been working on and what we are excited to bring to all of our customers to help them transform how they work with their content in this new era of AI.
Thank you, Aaron. In March, at Financial Analyst Day, we set out the vision to be the secure content platform for any AI-first enterprise. Since then, we have delivered extraordinary innovation with dozens of major releases. Customers will not let an agent work with their content unless the security they already trust still holds, and they do not want more security models. Agent guardrails now cover Box custom agents and MCP, controlling the actions they can take. Shield input safety, a full audit trail, and FedRAMP High+D give the certification that make using agents safer and more compliant. Customers also need the agent to be something the enterprise can rely on, which is why Box Agent has a pro mode for more difficult tasks, and agents can be called through an API or through MCP to make them available wherever your work happens.
To be even more valuable, context and agents should be available in workflows. We released Box Automate with advanced agent guardrails and integrate Box Agents into Box Apps, so teams can naturally work with agentic workflows. Context has to be available across the stack the company already runs, which is why we shipped Box for ServiceNow, Box Sign for Workday, and Box for Salesforce with Box MCP security guardrails and the action platform keeping the connection governed. A global company has to keep content in the region the business requires, which is why new Box Zones are now open in Singapore, Switzerland, and Israel as well. What we are announcing today carries the same vision farther. With Box Context Retrieval, we are bringing governed content and managed retrieval into one foundation.
Box already manages the source content, permissions, governance, and the version history. Box Context Retrieval adds automatically maintained indexing and retrieval, removing the need for a separate retrieval stack for your content. Integrating NVIDIA models expands the options customers have when selecting models for Box AI workflows. Nemotron embedding models supports retrieval, and the open-weight language models support generative AI use cases. Box Architect is our vision for how workflows get built. Instead of mapping the whole process before a workflow can start, Box Architect starts from the outcome you want. It reads what is already in Box, builds the workflow with no code, and recommends the next change. A bank and a studio do not work the same way, which is why Box for Industries starts from the way the industry already works, including FedRAMP High and GxP for life sciences as well.
Security has to cover the agent and not only the person, what an agent can reach and what an agent can do. We are extending threat protection and governance to quickly understand and respond to agent actions. Finally, Box Automated Archive, powered by Box Automate, will let admins manage the life of content across the whole company. As you can see, we are delivering on our vision to be the secure context platform for any AI-first enterprise. Now, I will hand over to the team to take us through these exciting developments, for which allow me to introduce Ben Kus, our Chief Technology Officer.
Thanks, Diego. Hi, I am Ben Kus, CTO of Box. As Aaron illustrated earlier, for agents to be successful, they need to understand your business, and that means having access to business context. That context is spread across multiple data platforms, in structured data stores, business applications, and unstructured content. It is that unstructured content that lives in files that creates a unique set of challenges for agents. This unstructured content not only has some of the largest volumes of data in your organization, it also is used by virtually everyone, and it is constantly changing, being created, edited, and shared. These files contain some of the most sensitive information in an organization, so security and governance are critical, and it comes in an enormous variety of formats that agents need to understand.
Because these agents are not storing or managing the data themselves, they need specialized tools to securely find and access this content. These are exactly the challenges that Box is built to solve. Box provides a single platform to securely manage your enterprise content, making it accessible to employees, developers, applications, and now to agents. The Box platform starts with virtually unlimited storage for your enterprise content. On top of that, we provide enterprise-grade security, compliance, and governance with granular permissions, auditing, and threat protection. We have a rich set of content services to organize, collaborate, and automate workflows around your content. Now, our agentic platform brings all these capabilities to agents, including integrated directly into Box agents, in addition to third-party agents like OpenAI, Anthropic, or any agent via multiple technologies like MCP, CLI, A2A, API, and more.
But there's a challenge that has existed as long as we've had enterprise content systems. How do you find the right information across all those files? Historically, that's been a search problem for end users, but with agents, it becomes much more critical. Agents need access to information that they're authorized to see, that's up-to-date, and that is actually relevant and accurate. Because if agents are working with the wrong context, they're going to produce the wrong results. Over the last few years, many enterprises have tried to solve this themselves, building their own retrieval systems using vector databases and retrieval-augmented generation. But they run into some major challenges. Traditional RAG approaches struggle to scale to millions of constantly changing files, especially when you factor in permissions. And preparing and parsing all those different file formats is incredibly complex.
Indexing everything upfront becomes prohibitively expensive in terms of infrastructure costs. Even if you get all that working, retrieval technologies are evolving so quickly that you're constantly having to go rebuild and optimize your system. That's why I'm excited to announce Box Context Retrieval. Built directly on the Box platform, Box Context Retrieval is designed to solve these challenges at enterprise scale, giving agents secure access to the right business context, powered by the latest retrieval technologies. Box Context Retrieval starts by handling the complex work of preparing your enterprise content for AI. It processes different file formats, including multimodal content, and it transforms them into searchable representations. It builds both the traditional text indexes and the vector embeddings. As files change, the indexes are automatically updated, so agents have access to the latest content.
When it's time to retrieve information, we combine the precision of lexical search with the power of semantic search, which understands meaning and intent. We then use a technique called rank fusion to combine and rank the results, giving agents the best of both approaches. So whether you're looking for an exact contract number or documents related to a business concept, we can find both. Importantly, we're not just retrieving the most relevant files, we're retrieving the most relevant information within those files. These capabilities are integrated directly into the Box Agent for agentic search and also available on third-party and custom agents through MCP. So any agent can leverage the power of Box Context Retrieval. The results are significant. In recall benchmarks, Box Context Retrieval improved accuracy by over 20%.
Because agents can retrieve the most relevant sections of documents rather than processing the entire file, we reduce token consumption by approximately 90%, which means that the agents can deliver answers four times faster. So agents get better context at a fraction of the cost and much faster. All this is integrated directly with Box permissions, so agents only retrieve the information that they're authorized to access. There's no need to build your own vector databases and indexing pipelines. Retrieval just works with the content in Box. As retrieval technologies continue to evolve, Box handles the updates, the re-indexing, the new techniques, so you don't have to. Of course, the quality of the context retrieval depends heavily on the underlying models. They need to deliver both the accuracy and the efficiency at enterprise scale.
That's why we partnered with NVIDIA to bring their latest Nemotron retrieval models into Box Context Retrieval. I'm also excited to announce that we're bringing NVIDIA's open-weight Nemotron models to our Box Agents, giving customers even more choice and efficient models. Putting this all together, Box provides a single platform to securely manage enterprise content, make it accessible to employees, developers, and agents. With that, I'll turn it over to KK to illustrate how we can bring these capabilities together to power advanced business processes across the enterprise.
Thanks, Ben. Today, I'm excited to show you how Box is innovating with agents and workflows to power the way teams work together in this new era of AI. With models getting better every day, the workflows will look completely different tomorrow. At their core, all major processes are centered around enterprise content. They are how teams onboard new clients, design new products, collaborate on contracts, and ultimately make business decisions. Let's start with the engine behind the workflow capability. First up, the Box Agent. Box Agent is purpose-built for unstructured content to find, understand, and act on enterprise content with secure access to customer data. It adaptively routes work, reasons across hundreds of thousands of documents, optimizes token usage, and supports long-running processes. It is designed to mitigate agentic security threats like prompt injection with guardrails.
With Box AI Studio, teams can create customized agents tailored to their specific business needs. The result is AI that can research, analyze, and work with complex content and your workflows. Today, we're announcing these new capabilities that will be coming soon to Box Agent. With safe web access, enterprises can search beyond their content to bring in relevant web information. With integrated skills, customers can teach the agent how they work through reusable skills and instructions, and keep working while they multitask across complex presentations, spreadsheets, and other content for up to 20 hours, and turn these analyses into deliverables like presentation spreadsheets and other files from their content and insights already in Box. As you know, Box Extract combines the latest AI models, advanced OCR, agentic extraction approaches to automatically and accurately extract structured data from content at scale.
Because Extract is built natively on Box, it has enterprise-grade security and permissions and governance controls. Today, we're announcing new capabilities that will be coming soon to make Box Extract even more intelligent, efficient, and trustworthy. With integrated eval sets, customers can measure extraction accuracy, and those eval sets then enable agents and cost optimization. Human-in-the-loop review applies custom rules to focus attention on potential issues. With adaptive model routing, it matches each workload with the right level of intelligence. Box Automate was designed from the ground up to orchestrate end-to-end processes across your people, AI agents, and enterprise systems. With Box Automate, you can integrate AI agents natively into workflows to perform both agentic and deterministic outcomes and scale automation to all of your business processes.
With Automate, available today, customers can customize agents for their business use cases, from models and instructions to the business knowledge base and outputs. Teams can add humans in the loop by assigning tasks to people on set criteria and confidence levels. They can apply advanced guardrails to set clear boundaries that restrict certain unwanted agent activity. Enterprise integrations connect Automate workflows to third-party applications across your entire ecosystem. Box Apps enables customers to build custom apps with a unified no-code builder. It powers views with metadata to surface key tasks related to your content, and agents work directly in apps to take on complex tasks. All of this allows you to manage your critical processes from one central place.
Our customers are using Box Apps to create tailored dashboards for the work they do every day, from contract lifecycle management, insurance claims, to customer onboarding and digital asset management. Frontier models have reached a level of intelligence where they bring immense global domain expertise to problem-solving. Imagine pairing those frontier intelligence directly with all the rich content, metadata, and security context already living inside your Box instance. Today, I am thrilled to announce Box Architect, an autonomous agentic system that partners with your enterprise to reimagine your workflows in the era of AI agents. Box Architect fundamentally changes how work gets built across your organization. First, it designs AI-first workflows with super intelligence, applying frontier model reasoning to your business goals to propose the optimal workflow. Second, it builds complete solutions autonomously, generating purposeful Box and Extract agents, structuring metadata templates, and building the Box Apps and workflows you need.
Third, it optimizes workflows recursively. As your workflows run across Box, it learns from your data, identifies bottlenecks, and refines the processes over time. This is more than building faster. It is reimagining how work runs. With Box Architect, you transform content-heavy operations into autonomous workflows that move your business forward. Soon, Box Architect will help you reimagine how every process runs with frontier intelligence. Today, these capabilities do not just help automate individual tasks. They enable enterprises to build, customize, and deploy agentic workflows at scale. Let us switch to how these solutions get deployed. Today, we are announcing Box for Industries. Box for Industries is Box tailored to the way industry works. This means workflows, metadata templates, certifications, and functionality built specifically for your industry. Workflows built around how teams actually work, like M&A due diligence at a bank or benefits processing at an agency.
Metadata extraction templates that pull the data right out of your documents. Certifications and compliance for the standards your regulators expect, like SEC 17a-4 or FedRAMP High. Functionality for each industry, such as our new customer onboarding solution for banks or the CAC and PIV smart card signer for government. Today, we are launching Box for Banking, Box for Public Sector, and Box for Media and Entertainment. Let us dive into the first two. Let us start with Box for Banking. Financial services firms use Box for everything from client onboarding, digital vaults, onboarding, policy management, due diligence, records management, to private credit. Next, Box for Public Sector, where agencies manage everything from case files, grants management, FOIA requests, evidence managements, all the way to benefits processing and records management. Every industry is different, and we are building Box to fit the way each one works.
This is just the beginning of the industries we'll be rolling out, and we cannot wait to build what's next with you. Next up, Manoj will talk about our latest security and governance updates. Thank you.
Thanks, Kelash. My name is Manoj Asnani, and I lead security and governance products at Box. Security has always been one of the most important pillars for the Box platform, and we've continued to enhance the way we protect organizations' content. Over the past year alone, we've delivered over 50 security capabilities across Box Core Security, Box Shield, and we've launched Box Shield Pro. As we are aware, AI agents are entering enterprises at scale. Organizations are going to have hundreds, if not thousands, of agents for every employee. At that scale, organizations need to distinguish trusted agent actions from activity that's malicious or misaligned while controlling agent actions. Consider recent incidents where a swarm of agents broke out of their sandboxes. They pursued objectives beyond what was intended. They escalated privileges, accessed credentials, and generated thousands of actions that are difficult to catch or track.
The risk isn't just that the agents can break out of their sandboxes, it's what they can actually reach and access, and that calls for a much stronger security approach. Organizations need to protect the content agents can reach and control the actions they can take. Because customers store, share, and collaborate on content in Box, we have access to a unique set of signals. Signals like collaboration patterns, like device posture, and content sensitivity or agent actions, and the data that's actually generated inside Box by user, app, and agents. A standalone point security solution may see an event. Only Box can connect multiple signals. At Box, we bring these signals together that help protect content and respond to threats. With Box Shield, we use collaboration patterns to surface anomalous activity such as suspicious sessions or anomalous downloads.
Box Shield Pro adds AI-powered understanding of content and supercharges classification of sensitive information. Box Core Security uses device and session context to help detect compromised logins or sessions. Now we're taking our security capabilities even further to protect content in the new era of AI. We're announcing security innovation on three fronts to protect organizations in this age of AI. We're supercharging threat protection, we're launching adaptive response capabilities, and we are providing security controls for agents. Let's first talk about all the innovation we're announcing on threat protection front. We're significantly reducing mean time to detection for activities such as anomalous downloads. We're strengthening protection against account takeover and session hijacking with session binding and device-bound session credentials, making stolen sessions harder to reuse. We're expanding anomaly detection for activities such as suspicious sharing and mass deletion.
Now, detecting a threat is only a part of the job. The response also needs to match the level of risk. We are announcing adaptive response capabilities for that. Adaptive response can prompt for stronger verification, limit downloads while allowing approved viewing, or end a session altogether. That helps teams respond proportionately instead of treating every event the same way. On the agent security front, we are launching four key capabilities. Box Shield classification-based access policies apply content rules before an agent can access a file. MCP guardrails lets admin set boundaries on agent actions and tools, including creating, moving, and sharing content externally. Threshold-based policies alert teams when selected activity crosses the configured limit to protect against malicious agents causing massive damage. Finally, agent session governance and audit trails show which agent acted, what content it touched, and the output it generated.
Together, these controls help keep agent activity within policy, so organizations, even in the most regulated industries, can put AI to work while keeping agents within policy. Protection is only one part of the story. Organizations also need to manage content responsibly through its full lifecycle. Box Governance helps organizations manage content throughout its lifecycle. Admins can apply retention policies, preserve content with legal holds, and manage disposition rules with auditability throughout the process. As more and more people and AI agents create content, organizations face a growing challenge with content proliferation. How to manage this content across its lifecycle? Organizations need an approach to governing records and content across the enterprise. Today, we are announcing Box Records to help organizations manage and retain their content as records where they already live, in Box.
Enterprises can identify, classify, and track official records across folders and systems, apply retention consistently, automate record declaration, find records for audits or legal needs, and manage disposition with confidence. As content and AI-driven work grows, manual archiving becomes harder to sustain. Admins need a consistent way to archive content at scale. That is why today, I am excited to announce Box Automated Archive. This is a new capability powered by Box Automate that gives admins an automated way to manage content lifecycle across the entire organization. Admins define rules based on real business events, like a project marked closed, and when those conditions are met, archival happens automatically. The result is a stronger compliance posture and cleaner workspaces. Admins go from manually enforcing retention policies to simply defining the rules, and Box does the rest. With Box Automated Archive, we are giving admins the tools to enforce governance at scale.
Together with Box Records, these capabilities give admins ways to automate governance, control over records throughout their lifecycle. This is Intelligent Content Management. Box brings content, AI, workflow, and security together, helping organizations turn enterprise knowledge into business momentum. Thank you. Back to you, Aaron.
Thanks, Manoj, and thank you to the rest of the product leadership team for the amazing innovation that you have been driving and in partnership with all of our customers. As you saw, and as you have seen from the rest of BoxWorks, we are driving the most significant transformation of our platform that we have ever seen. We have been investing in really the next breakthroughs in how do we bring the full power of AI to our Intelligent Content Management platform. We are on a journey to help our customers with this transformation. As you see the innovation that we have been driving, we obviously want to continue to put a core emphasis on helping all of our customers take advantage of the power of AI within their Box environment. But we are uniquely focused on our Enterprise Advanced plan.
We believe that this really is the frontier plan within Box that will contain our most advanced agentic capabilities. So, you saw the Box Architect, which is now an autonomous system to help our customers build, design, deploy, and eventually autonomously optimize their workflows with enterprise content. So you can almost think about it as the world's most advanced engineer that knows everything there is to know about enterprise content and Intelligent Content Management. That will be one of the staple capabilities within the Enterprise Advanced plan. We also have Box Context Retrieval, which makes both the Box Agents as well as external agents even more effective and token efficient at being able to retrieve data and answer questions from documents. So that is a core part of the Enterprise Advanced plan.
And we have introduced all-new threat detection and AI security features to help our customers protect and govern the activities that their agents are doing, again, whether those are Box Agents or external agents. And a number of those core capabilities are also within Enterprise Advanced. So we are helping our customers on this journey, and we are continuing to focus on how do we help our customers move up into Enterprise Advanced by delivering an incredible amount of value for them on this transformation journey. So, we are really excited again about the BoxWorks innovation that we are driving. We actually have a lot more in store that we were not able to cover even at the event, so stay tuned for even more announcements in the coming weeks, months, and quarters. And with that, we will open it up for questions.
Great. Thank you. Go ahead. Great. Thank you, everyone. We will take our first question from the audience.
Thanks for doing this, guys. Great presentation today. Lots to take in. Lucky Schreiner with D.A. Davidson. Maybe just starting out on the NVIDIA partnership, can you elaborate on how that differentiates your Box Context Retrieval versus some competing enterprise search and RAG solutions, and specifically how defensible you view that advantage going forward? Thanks.
Yeah, great. Let me I'll characterize it for five seconds, and then maybe, Ben, if you want to get into some of the technical side on this. We think about it as, first of all, generally complementary. Customers can use whichever systems they want to be able to have federated indexes and whatnot. There's a fundamental problem, which is, if you just wanted to make sure that you had the best context for your agents from enterprise content, you don't want to have to go and build out an entire stack. So what we wanted to do is make it very, very easy for you to do this directly in Box, basically with just an API call or MCP call or any agent that is connecting to Box. We wouldn't necessarily position it in any kind of competitive way with any other solution that you might go do.
It's just a different approach if you want to have an out-of-the-box way of delivering this within our platform. Then we will connect up to any other environment that you have for agent orchestration or other ways of federating in that context. This will be the best way to get context from your enterprise content, and that is the Because we're going to go deeper in just the enterprise content set of use cases. That's the sort of value proposition of Box, but we wouldn't position this as just trying to go after that market and compete discretely with those players.
Yeah. A lot of our customers, they had, in many cases, said that they really needed Box Context Retrieval to power their agents going forward. But one of the challenges is when they try to do it, they would say, "Oh, I could get a vector database. I could do it myself." Then you just run into no end of problems. Most enterprises don't have the resources to invest in, especially since it's actually quite expensive from an infrastructure perspective to be able to do this kind of work. You typically spend a lot of time optimizing it to making it work well. So to the point about the Nemotron models, there's relatively recently a new set of embeddings models that came out from different vendors, including from NVIDIA has one of the best ones, the Nemotron retrieval models.
This allows the agents in the context retrieval system to get much better results. The semantic vector embedding sort of comparison, it just works better these days. It's similar to how the new models are better than they used to be. They're smarter and so on. The embeddings models are just kind of better at matching this kind of, especially with the business context. Many customers are looking at this and saying, not only that they want to use the latest, but then also realizing that these come out a lot. Over time, there's these newer and better models, in which case it's actually hard to go back and re-index everything. That's why we got a lot of customers asking us over time to say, they're sick of maintaining their own retrieval infrastructure.
In particular, because they also want to match the permissions, and that's just very hard to do, especially as people are sharing stuff and moving files around and so on. Combining it all together, the Box Context Retrieval built directly in Box is something that is one of the main things that many of our customers are asking for where they store their data. To Aaron's point, it's because we store the data, we store the permissions, we have all that ready, Box Context Retrieval is a natural byproduct of that.
Thank you. We also have a question. How does Box Architect differentiate from other workflow automation and process orchestration platforms?
Yeah. Architect is actually a layer above the workflow platform. Architect is an autonomous builder of the workflow, the business application, the data extraction process, even managing the underlying sort of file system beneath it. The best way to think about it is, what if you took the best of frontier intelligence, Astra, Fable, et cetera, level models, and what if instead of just applying them to steps in a workflow where they process a document or do data extraction or answer a query inside of a question from an agent, what if they actually designed the entire process, the client onboarding process, the contract review process, the invoice processing steps?
Now you have frontier intelligence actually help you figure out what is the most optimal way you should design your workflow, then it builds that out with Box Automate, or Box Apps, or Box Extract, whatever the underlying tools within Box are that need to build that, as well as eventually connecting to other systems that it might need to coordinate with, that agents need to be able to interact with. So, in an interesting way, there is probably almost nothing like this on the market. It is much closer to if there was an autonomous system that basically did your entire enterprise content workflow, and it is much more akin to the types of things that, in our case, our forward deployed engineer or a technical consultant would do in your organization. So it does not really replace a piece of technology.
It is much more about bringing you a level of automation and intelligence for powering your entire end-to-end process. It is actually something that could not have really existed even three months ago. The models were not good enough at being able to do that level of planning and tool use and choose the right agent and choose the right model for the agent, and hill climb on a data extraction pipeline. No, the models just would have been ineffective at that. So, brand new category of how we start to think about autonomous systems at a little bit more of a meta layer of now the autonomous system is running the process to begin with and helping us deploy, run, design, and then eventually optimize. But yeah, breakthrough new capability.
Yes. It is like an outer loop that connects and stitches all the other products and can design or optimize an existing design of the whole workflow. It is applying intelligence on workflows.
Thank you. We have a question regarding the Box for Industries announcement, and how does this Box for Industries and these segments you rolled out compare against purpose-built industry-specific content and compliance platforms?
Yeah. We spend a significant amount of our time really building out the leading Intelligent Content Management platform, emphasis on the word platform, which means all of the core capabilities on security, governance, access controls, compliance, regulatory features that you need, data governance, all of that. Then we have a layer on top which does our content services for content management, workflow automation, our metadata system, all of that. Then obviously a very significant agent harness and AI platform. Our customers end up deploying that in more business-specific ways. They obviously don't just take us in as a horizontal platform and then use us horizontally.
They take us in as a platform and they say, "Okay, I want to do clinical trial processes in life sciences," or, "I want to do contract management at a law firm or a commercial real estate organization," or, "I want to do client onboarding at a wealth management company or wealth management firm." The way that Box already gets deployed is in this sort of more vertically oriented way because that's the use cases of the organization. What we are trying to do is move up the stack in terms of being able to deliver those solutions far faster to our customers, at least when they are sort of heavily dominated by the core Box primitives.
If you have a very content-centric process for client onboarding, loan processing, contract review and approval, any of those types of business processes, we want to make it insanely easy and fast to go and get that solution built for your industry, which means we need to have some opinions about how that would get deployed, what partners we'd bring in, how you'd want to roll this out, how you should design the solution. Box for Industries is basically us bringing that level of expertise and domain specificity to your workflow, then supercharged by capabilities like Box Architect, where all of a sudden, it's not just the Box team and Box partner ecosystem, it's actually frontier AI helping us design that solution for my workflow. Then this will also naturally expand into the broader partner ecosystem.
Just yesterday, we had our partner summit with the leading systems integrators and GSIs, boutique SIs, and hyperscalers around really how do we bring the Box platform to our customers in this more vertically oriented, solution-oriented way. You're going to see Box for Industries also bring along the partner community. One just kind of finite example of that is, we do a lot in the insurance space with our partner Guidewire, and Guidewire is a leading insurance claims kind of processing system and SaaS platform. Every single insurance claim deals with a lot of unstructured data, video, documents, information from the insurer, information from the ultimate consumer, and that all has to be stored, recorded, governed, tied to a workflow.
We are increasingly an intelligent content layer tied to Guidewire, brought in in addition by systems integrators, and now increasingly more agentic because of things like our data extraction agent, and now ultimately Architect designing that full solution. There is going to be both Box-specific products as well as an industry partner angle that will get Box more embedded into these deeper workflows.
Great. Thank you. Another investor question we have coming in from online is, how do Box's, your newly announced threat protection MCP guardrails, compare against specialized cybersecurity vendors that offer centralized AI security across an enterprise entire SaaS estate?
Do you want to?
Our implementation is very connected to the and being built in from the ground up within the platform, with the underpinnings of our own platform itself. That is different from having external solutions protect content, and also is very tied to governance. We have security and governance very connected with each other, where you can select which files or which specific content can be accessed by whom, including agents, and basically connecting security with governance and maintaining that also tied to compliance makes it very unique to the Box platform.
Yeah, I mean, on the platform layer that Diego just mentioned, so we have, and this was covered for a couple seconds, but it is hard to pack in. We had probably like five hours of content for an hour and a half keynote, and then the investor presentation is obviously even more condensed. Our Shield technology takes in an insane amount of signals from basically every single activity that happens on our platform. Every time a file is viewed, every time it is downloaded, every time you add something to a file, every device it is touched, every application that has access to it, that is a level of intelligence that is just very hard to replicate off-platform.
That means that we can offer a unique perspective in how do you secure data, and the flow of that data as agents are coming in, as applications are leveraging the data, as users need to access that. So you can detect things like anomalies, or you can detect, these are strange patterns that we have not seen before. You also become a very obvious sort of gateway for external agents that need to route into via MCP enterprise content. Instead of having customers have this many-to-many relationship of many agents and many content systems, they end up, no matter what, they are going to have many agents. That is an inevitability.
But they want to be able to have some kind of choke point or control plane that says, "Okay, these agents can access this type of information." If you are in a healthcare provider, they do not want to be able to access HIPAA data, but they do want to be able to access maybe internal procedures that need to be maintained by healthcare professionals. Thus, you would need to have data classification that says, "This is your HIPAA environment." All of that cannot really be done very easily in a very, very fragmented environment. So it is causing customers to actually move more of their content into our platform, and then the need to then have guardrails and security and compliance on how do I protect that information.
So we become a very natural area to do a lot of this advanced AI security, because the content is there, the classification needs to happen there, and then we have all these sort of extra signal that a lot of other platforms do not have. So we think that you are going to see, I mean, we know you are going to see a lot more from us on the security front because of how important this is around routing agents to the right information. So stay tuned. Today's announcements are truly just the start of what you are going to see on the cyber front.
Great. Lucky Schreiner again with D.A. Davidson.
Good to have you back.
Yeah. Just back on Box Context Retrieval and Box Architect, how do you plan to bundle that and integrate that in the broader Box AI Studio experience and just general monetization plans for those products?
Yeah, so both of those products are in the Enterprise Advanced plan. The rough sketch of a lot of what you saw today is brand new additions or advancements of our Enterprise Advanced functionality. We're still committed to deeply investing in the core Box products that will kind of benefit all of our plans. But a lot of our most advanced agentic work, a lot of our most advanced security work, especially dealing with agents and Box Context Retrieval to prepare data for agents, they really kind of logically make sense to be packaged in the Enterprise Advanced plan because that's kind of where we are bringing the frontier of our AI solutions and our AI capabilities. So we think this provides another set of catalysts for customers to upgrade into Enterprise Advanced. It gives them even more value.
Again, whether or not the agents are being designed in Box in our AI Studio or it is an external agent, things like Box Context Retrieval benefit both of those ecosystems deeply.
Hi, Rafi from Emmett. I just want to make sure I show up in the transcript again.
Yeah, exactly.
That is the only reason I am here. For Box Architect, because I am just trying to make sure I understand this stuff. I am kind of slow on the technology stuff. Is it one of those things, like, not that you guys are competing with Palantir or something, but is it one of those things that, like, as you get Box Architect going, there is a snowball effect where, like, the first workflow is good, but then by the 10th, it is even better. Like, it builds on itself. Is that how to think about it, or is everything just a one-off? Kind of like, yeah, does it snowball effectively for an organization as they get into this?
I think one of the big values is that, when we talk to a lot of our customers, they say that one of the things they're inherently limited by is their internal capacity to improve their processes, because everything's changing so rapidly, and they're trying to use the latest capabilities. But at the same time, it takes somebody's time to sit down and understand them, and then try to map it to a process of how even agents work. Then they would oftentimes come to us and say, "Okay, can we have some of your forward deployed engineers?" But even that is a process that they'd have to invest in and work with our FDEs.
Box Architect, one of the key things is that it just makes it very easy for you to have that kind of expert level support with, I mean, obviously, these days, the newest models, they can do very sophisticated programming. They can make applications. So applying that same level of intelligence to the capabilities that we provide, and then having the customer give inputs with their data so that then it can create the process, and they can start just to be able to get all these benefits very quickly. Some of the early examples that we're seeing with early customers we're working with on this is that they're looking for a lot of breadth.
They're like, "Oh, if I can get that working, previously I had planned to do this amount of, I'm going to automate these kind of processes in this amount of time." But they're like, "Well, if it's that easy, and if I can then apply that repeatedly, then I can do more and more and more." So we see more examples and more breadth as one of the key sort of benefits of Architect.
Yeah, I mean, if you Or sorry, go.
Thank you. One important point is that as intelligence continues to advance and you get super intelligence or closer to it's conceivable that the workflow that you had built before can be improved. On top of that, some of the components of the workflow might be changing, so you may want to end up changing the whole workflow because of the components have advanced, or the process in the industry or in your particular enterprise might change too. It's important to have a way to continuously improve it, not because it's designed to be changing all the time, but because the nature of the business may be changing or the nature of the components.
Having a system like Box Architect that first can help you design it at the beginning when you get started, but also to adaptively improve as components or the whole system change is essential. It's not nice to have, but we think it's going to be something widely adopted across the industry.
Maybe just I'll build on one final point, just to make it really practical. For something like, let's say, data extraction in a workflow, whether you implement the first model you choose and the first sort of tuning of kind of explaining that model what you want to do for extracting your documents, versus do you put in the time, the 10, 20, 50 hours of tuning every last change of a model, every last change of how do you pre-process the data, that could be the difference between an 80% extraction rate and a 99% extraction rate, or kind of assume putting it 100 or 99.99. But to Ben Kus's point, you rarely have the talent internally in your organization to go do that, especially once you get meaningfully outside of the tech industry. What are the engineers in Silicon Valley doing at this point?
They're using AI to do that. If you go talk to our agent team that builds agents, they don't handcraft the perfect prompt that an agent uses. They use the frontier AI to help them run the eval sets to then go and design the perfect agent to execute whatever the work is we're doing. Imagine if you could bring that kind of style of automation and autonomous systems to any process in an enterprise that deals with enterprise content, because that's obviously what we're focused on. Now you come to us. This is more philosophical. It'll be live in the coming months. You come to us and you say, "Here are my contracts.
I'd like to achieve 95%, 98%, 99% sort of accuracy at this kind of rough cost profile based on what I'm trying to do. Architect will use the internal tools in Box to run the extraction processes with your internal eval sets that you give us or ground truth that you give us, and iterate over as many sort of trial runs as it needs to find the perfect optimization of, how do I deliver that accuracy for you for this type of data extraction? That's like one of 100 use cases on any given enterprise. It could be any type of business process that a customer can think of. This is, again, closely resembling what if we achieved autonomous content management and content workflows? You would need something like Box Architect to go and deliver that.
We're just introducing today the kind of first phase, which is the building. But when you have the building, the running, the tuning, the optimization, it sort of completes the full cycle.
Great. Well, we're at time, but Aaron, I'll let you wrap it up.
Great. I mean, first of all, appreciate everybody tuning in. If you get a chance offline, because we couldn't cover it in the IR, please watch the full breadth of our keynote. There's also kind of a great intermission section with Jensen Huang laying out, I think, not only our partnership, but really the opportunity that enterprises have with unstructured data. It's so unbelievably clear to us that there's a very different moment right now around the role and power of unstructured data in the enterprise. It's causing a complete shift in how companies are thinking about how important this asset is in their organizations.
When you looked at, if you kind of reverse engineer, like, what would you then need to give agents a secure container that they could work with, that other people could contribute data into, that has the access controls and has the cyber capabilities? You would eventually end up with what we are building, and we are the only company that has designed that full end-to-end platform in the cloud in an AI-first way. We are really excited about the position we are in. We are very excited to bring this to all of our customers, and we will certainly keep everybody posted on the roadmap from here. Thank you. Right.
We are cut. Thank you.
Thank you.
Thank you.