Box Unveils New AI Agents for Enhanced Enterprise Content Management

Box's new AI agents can organize, find, and extract data from documents for you

They will be released in the coming months for enterprise customers. AI agents, as you've probably noticed, are all the rage in Silicon Valley. On Thursday, the content management platform Box joined a growing list of companies hoping to cash in on this latest tech trend. The new Box AI Agents are designed to help enterprise customers organize and retrieve critical information from files across the platform. Also: 100 leading AI scientists map route to more 'trustworthy, reliable, secure' AI Like many new "agentic" products, the agents are promoted as time-saving tools that enterprise customers can harness to reduce mundane tasks that tend to eat up large chunks of employees' workdays, like summarizing HR forms or pulling key details from lengthy contracts. The agents are being released as part of Box AI, the company's AI-powered content management tool, which debuted in late 2023. Also: Tech leaders are rushing to deploy agentic AI, study shows Box also announced the launch of an AI agent that can be integrated with Microsoft Copilot to access and move between platforms like Teams, Word, and PowerPoint. This cross-functionality is another key selling point for Box and companies pushing agents more broadly. "The future of enterprise AI will be defined by intelligent agents that can work together across systems, each bringing unique context and capabilities to the table," Aaron Levie, cofounder and CEO of Box, said in the release. "Just as APIs once connected software, AI agents will change the way we work -- and that transformation will be most profound when applied to enterprise content." Agents are to more conventional chatbots (like ChatGPT) as a football team is to a single linebacker; the former consists of several integrated components, each acting toward a unified goal. An "agent architecture," as it's known in the field, enables these systems to retrieve information, formulate strategies, and autonomously take action on behalf of users; think of it like the playbook that dictates how a football team operates on the field. Also: Worried about ChatGPT saving your chat data? Don't be - just click this one button Box AI Agents, for example, use a collection of AI models from Amazon, Anthropic, Google, IBM, Meta, OpenAI, and xAI. The system then adaptively pairs users with one specific model, depending on the nature of the query. Overall, the new agents are designed to make it easier for enterprise customers to sift through mountains of documents, much of which is siloed by division, client, and so on. Also: Microsoft 365 Copilot's two new AI agents can speed up your workflow "At the core of the new Box AI platform is a dynamic agentic reasoning framework designed to make sense of today's complex content landscape," the company wrote in the press release. "Box AI Agents operate where content lives." The new Box AI agents will be released in the coming months, along with pricing details, according to the company. Get the morning's top stories in your inbox each day with our Tech Today newsletter.

Box tightens Microsoft ties with new Copilot integration, builds out its own suite of AI agents

Box is deepening its connection to Microsoft's productivity platform through a new AI agent that works with Microsoft 365 Copilot, while introducing new AI agents of its own to analyze documents, automate tasks, and dig up insights from enterprise data. The new technologies, announced Thursday morning, build on the Redwood City, Calif., company's longtime role as a hub for secure file storage and collaboration, while continuing its push into what Box describes as intelligent content management. Microsoft 365 Copilot is Microsoft's AI assistant for business, integrated with its widely used productivity apps. The new Box AI agent will let users search, summarize, and act on content stored in Box from inside Microsoft tools such as Teams, Word, and PowerPoint. Box CEO and co-founder Aaron Levie, who grew up near Microsoft in Mercer Island, Wash., said in an interview that the Copilot integration is part of an effort to make data and content in Box available to AI applications in other environments, in addition to its own platform. "We want to be the best place for bringing AI to content, but we know there's lots of other things you're going to do with AI," Levie said. For example, he said, a user might ask Microsoft Copilot a question that pulls information from tools like Outlook or Dynamics. Box wants to ensure content stored on its platform can be part of that answer, alongside other enterprise data. In addition to the Microsoft 365 Copilot integration, Box is introducing three new AI agents on its platform, expanding on its existing AI capabilities. The agents focus on intelligent search, deep research across large sets of documents, and structured data extraction from unstructured sources like PDFs and scanned images. Levie has emerged as an outspoken advocate for using AI tools in daily work, writing in a recent LinkedIn post that he now turns to AI for everything from drafting product ideas and customer surveys to researching market trends and analyzing investor materials. The industry's move into AI agents aims to take things a step further. The idea is to not only generate content or summarize information, but complete tasks across different systems, like pulling data, connecting tools, and doing other types of work automatically. Speaking with GeekWire this week, Levie said AI agents have a lot of promise for automating knowledge work, but acknowledged that the technology is still early, and easy to overhype. He described 2025 as largely a year of pilot projects and smaller-scale testing, and said many ideas for agents are not realistic yet in practice. "If you ask 10 people to describe AI agents," he said, "you'd probably get 20x the number of use cases than are probably possible today." At the same time, he said, some use cases for Box's new AI agents are already proving valuable, citing two examples. In both cases, Levie said, the agents are enabling initiatives that previously would have been too costly or time-consuming to justify. While Box and Microsoft compete in some areas, such as cloud storage and collaboration, Levie said they continue to partner across a range of areas. That pattern plays out across Box's business and the industry. Box also competes with companies including Google (Drive), OpenText (Documentum), and Dropbox. It also partners and integrates with Google, Salesforce, Zoom, Adobe, and ServiceNow. Box, founded in 2005, went public in 2015 and has about 2,800 employees. The company reported $1.09 billion in revenue for the fiscal year ended in January, up 5% from the year before. Profits for the year were $244.6 million, up from $129 million the year before. The company said its new AI agents will be available in the coming months, with pricing to be announced closer to launch.

Box expands its toolbox with more AI agents for search, deep research and data extraction - SiliconANGLE

Box expands its toolbox with more AI agents for search, deep research and data extraction Cloud content management giant Box Inc. today is expanding its artificial intelligence capabilities with the launch of new AI agents for search, deep research and data extraction that will automate yet more tasks on behalf of users. In addition, there's a new Box AI Agent for Microsoft 365 Copilot, which expands the reach of Box content into applications such as Word, Teams, PowerPoint and Copilot Chat. Box announced its latest AI agents at its virtual Content + AI Summit today. They build on the original agentic AI capabilities introduced in its platform in February. The company has emerged as a pioneer in AI, making the technology a central aspect of its content management platform. AI has been integrated into almost every aspect of Box, from its core metadata layer to the user experience. It began in 2023, early in the generative AI boom, when it announced a suite of tools called Box AI, aimed at helping workers to become more productive. Since then, Box has progressively updated its AI features, launching Box AI Studio last year to provide users with a way to create and use their own AI agents to perform increasingly complex tasks. It has also launched Box Apps, which is a no-code development tool for users to create intelligent applications and workflows for business processes such as invoice processing and contract management. Then in February, it launched its first AI agents to automate tasks such as querying documents and extracting data from files, providing faster access to the insights locked within them. With today's update, Box says it's about to enhance its agentic AI capabilities further, with new agents for search, deep research and enhanced data extraction. Box said the AI search agent (pictured) can perform "quick lookups," rapidly surfacing targeted answers such as an expiration date in a contract or the name of a client in meeting notes. It can also handle more complex queries, using high-precision semantic analysis techniques to uncover hidden relationships and insights from large volumes of documents. Speaking to SiliconANGLE, Box Chief Executive Aaron Levie said AI agents build on the capabilities of generative AI chatbots that have already become quite widespread among enterprise workers. "Unlike when you just ask an AI model a question, an agent will go off and do a bunch of work for you," he explained. "Maybe it uses a search tool, so it does a search query. It will read through those results, read the content of that document, and connect the dots with other documents." The deep research agents do something similar, but they're more focused, with the ability to analyze thousands of documents and files to surface meaning and trends. When given a task, they'll first identify the most relevant files using Box's retrieval-augmented generation framework, before synthesizing what it finds into digestible chunks, so users can quickly access the information they need. If you want to acquire a company, for example, Deep Research will help you to find everything on it, Levie explained. As for the enhanced data extraction, it will help Box's AI agents to transform unstructured information such as scanned PDFs and handwritten notes into structured data that's more actionable, the company said. They'll leverage technologies such as optical character recognition and document intelligence to find and extract key data, such as financial figures, contractual clauses and terms, dates and so on. Levie said Box is bringing its enhanced data extraction capabilities to partners such as Salesforce too. As an example, he said, if you're inside Agentforce and you want to find meeting notes about a client, the agent can pull that out. According to Box, its agents are powered by "best-in-class" AI models from companies including OpenAI, Amazon Web Services Inc., Anthropic PBC, Google LLC, IBM Corp., xAI Corp. and Meta Platforms Inc. This means users, or the agents themselves, can choose the most appropriate model for each task. Box said the search, deep research and enhanced data extraction agents will be available in the coming months, but customers won't have to wait to access the new AI agent for Microsoft 365 Copilot, which is launching in beta today. The idea with this agent is to link Box and its content management system directly to Microsoft applications such as Word, Teams, PowerPoint and Copilot Chat. When using those apps, they'll be able to ask the agent to retrieve content directly from Box, without actually having to access Box themselves. The agents will be able to search, analyze and act on that information. Levie said this is in line with the company's ultimate ambition to build an "intelligent content management platform" that will be able to bring users' content to wherever they're currently working. He added that the Microsoft integration expands a growing ecosystem of AI partnerships. Previously, Box has integrated its content and agents with platforms such as Google Agentspace, IBM watsonx Orchestrate, Salesforce Agentforce, Slack AI, ServiceNow AI Agent Fabric and Zoom AI Companion. For now, Box is focused on providing basic agentic AI features such as search and extraction, but it ultimately wants to build on these features, Levie said. "This is going to be a multiyear journey," he said. "Companies are going to take years to figure out how to use agents in their business, so we're still in an exponential part of the curve." As AI agents become more integrated with enterprise workflows, businesses are going to need to pay a lot more attention to their data architectures, Levie predicted. That's because they'll want to ensure that their unstructured content, or their CRM data, isn't stored somewhere where it cannot be accessed by AI. "You have a risk of not capturing the full upside of AI if you don't have a clean data architecture," he said. "We know exactly where our customer data is, we know where our financial data is, we know our content is in Box. You need to have these very clear swim lanes, and allow AI agents to have access to all of this data." Amy Machado, an analyst with International Data Corp., said Box promises to deliver a real breakthrough by enabling AI to understand and act on the vast amounts of content that are essential to enterprise operations. "Customers will be able to leverage AI to instantly find answers, extract insights, and create new content with their most critical information already securely stored in Box," she said. "It's a big and important step forward in making AI truly useful across the enterprise."

Box brings AI agents to enterprise content

Content management vendor Box yesterday launched its first AI agents to help enterprises search, manage and work with their content stores. The new agents provide the ability to perform targeted enterprise search, carry out research at scale to provide in-depth analysis of key trends and information, and automatically extract key information such as dates, financial figures, clause types, and terms from documents and other unstructured content, including scanned PDFs, images and handwritten forms. These new agents build on existing capabilities in the Box stack, with the addition of a reasoning engine to co-ordinate orchestration and apply the most appropriate models, while Retrieval Augmented Generation (RAG) helps direct agents to work with the right source material. Recent additions to Box technology, including last year's acquisitions of Alphamoon, which uses AI to automatically identify metadata within unstructured content, and Crooze, a no-code business process app builder, have added important building blocks that support these new agents. Speaking as part of a virtual event to launch the new agents, Aaron Levie, CEO of Box, was enthusiastic about the impact agent-powered automation will have on the ability to harness information that's been locked away in enterprise content stores. He explains: We've never been able to fully tap into the value of this information until now. AI agents fundamentally will let us transform the value of our content. For the first time ever, we can finally do things to all of our unstructured data that we've always been able to do with our structured information. We can ask AI agents to look across one or 100 or 10,000 or a million documents, to generate instant insights about our information. We can have AI agents go read all of our documents for information, to extract the most important metadata or structured data from those documents... And then, once you have structured data from all of your unstructured content, you can then begin to automate any workflow. You can move content through a process in an automated fashion, whether that's for approving a loan, onboarding new clients processing a research claim, getting a plant audit committed, or one of tens of thousands of other types of work. He also emphasizes the value of information that exists in enterprise content, but which organizations have not previously been able to access, because of the difficulty of extracting and classifying it from documents and other types of content such as images, audio and video. He goes on: For the 90% of our information that is unstructured -- our documents, our contracts, our media assets, our marketing materials -- all of that, we've never really been able to tap into. We've never really been able to look across all of that information to understand what we have and what we're working with. And if you think about it, this is, in many cases, our most important data in an organization. It's the data about our next great product breakthrough that lives inside of a product roadmap, or research information from our customers, or customer satisfaction information that we then use to turn into an amazing new product insight. It's the data that lets us go and recruit and find amazing employees, or where we onboard them make sure that they're trained with the right assets and information to execute on their work. It's the data that goes into a clinical drug trial that is going to have the next great cancer breakthrough in treatment, all of the research information and genetic data that we work with to drive these types of discoveries. Box is also conscious of the need to connect its agents into other application platforms. Yesterday, it introduced a new AI Agent for Microsoft 365 Copilot, giving users the ability to securely search, analyze, and act on Box content directly within Microsoft tools such as Teams, Word and PowerPoint. Other integrations are available or in development with Google Agentspace, IBM watsonx Orchestrate, Salesforce Agentforce, Slack AI, ServiceNow AI Agent Fabric, and Zoom AI Companion, while developers can add their own integrations using the new Box MCP server. The Box AI Studio, unveiled at last year's BoxWorks to allow developers to build their own agents with their chosen LLM and custom prompts, can be used to customize the new agents annnounced yesterday. Box's decision to support connections to a wide range of LLMs allows it to take advantage of the rapid evolution in the market, as Levie explained to Wall Street analysts in the vendor's Q4 earnings call last month. He told them: Just in the past two weeks, we've made new models like GPT-4.5 and Claude 3.7 Sonnet available to customers with access to the Box AI Studio. Both of these new models have shown dramatic improvements on reasoning, math, and coding skills, which is extremely valuable in processing complex documents contracts, agreements financial records, and more. In fact, in our Box AI enterprise eval, we found that GPT-4.5 performs a whopping 20 points better than GPT-4 did for extracting data from documents in a single shot. The new agents are slated to roll out to customers in the coming months, at which point pricing will be revealed. Box recently introduced an Enterprise Advanced subscription that includes all existing capabilities and will likely also include the new agents. Of note, however, is that Box has also introduced a consumption-based element to its AI pricing, which means customers will pay more when carrying out AI-intensive tasks. As Levie explained on the earnings call: We've added an AI consumption-oriented model to our platform with the launch of Box AI units. These AI units allow customers to leverage AI credits for high-volume tasks, like metadata extraction from documents, leveraging more advanced AI models as they get released, creating custom AI agents to automate workflows in the future, and more. An important step for Box as it joins the agentic AI trend -- but it's an interesting approach that contrasts with other vendors. A more typical approach is to work with a preferred set of LLM providers -- and sometimes custom models of their own -- rather than choosing to adopt a wide range of LLMs and apply whichever works best, or is mandated by the customer, for each individual task. The openness to connect into a range of teamwork platforms and interconnect agents using both MCP and other protocols is testament to a determination at Box to take a neutral, Switzerland-style approach as it applies AI to enterprise content.