June 2026 Summaries

At BoxWorks London, Box announced the expansion of its Box Zones, introducing three new regions—Switzerland, Singapore, and Israel—and enhancing in-region processing capabilities for France and Canada, reflecting the growing importance of data residency in global enterprises. This development caters to the specific needs of industries like financial services, healthcare, and public sectors where data residency has become a critical requirement due to regulatory demands. The expansion to 10 global locations aims to provide enterprises with secure, locally-stored data while maintaining the seamless collaboration and productivity Box is known for. Administrators can manage Zone assignments through various methods, ensuring that user data aligns with regulatory requirements without disrupting collaboration. Additionally, Box plans to introduce in-region storage and processing for custom metadata, Box AI, workflow, and search, aligning with evolving data residency and AI governance needs. This initiative underscores Box's commitment to offering secure, compliant, and efficient content management solutions across different regions worldwide.

The "State of AI in the Enterprise" report highlights that while 99% of organizations now incorporate AI, the value and return on investment vary significantly, with only 50% of leading-edge organizations reporting significant ROI. The report, based on a survey of 1,640 IT decision-makers across four countries, underscores three key foundations that set AI leaders apart: integrating AI with company-specific content, establishing effective governance frameworks, and maintaining flexible architectures to avoid vendor lock-in. It identifies a major challenge in making enterprise knowledge accessible and trustworthy for AI, with most organizations facing more of a content issue than an AI problem. Governance frameworks have advanced, but they still pose deployment challenges, as only a fraction of organizations have comprehensive visibility over AI use. The report also notes a significant shift in workforce dynamics, with roles like AI agent operators and governance professionals emerging rapidly, indicating a reshaping of capabilities rather than a reduction in headcount. In the EMEA region, the deployment of AI is particularly influenced by regulatory complexities and the strategic need for architecture flexibility to avoid vendor dependency, emphasizing the importance of knowledge control and permission governance.

AI adoption is widespread, with 99% of organizations using it in some form, but only a fraction are successfully harnessing its full potential due to a content-related gap rather than technological limitations. A study commissioned by Box highlights that UK enterprises, although slightly below the global average for integrating AI with trusted internal content, are well-positioned to improve given their higher than average business impact from AI projects. Organizations effectively leveraging AI have established robust governance frameworks and integrated AI into enterprise knowledge systems, enabling them to scale and achieve significant ROI. In the UK, regulatory challenges, security, and data privacy concerns are prominent barriers, yet AI is expected to create new roles rather than eliminate jobs, as many organizations anticipate an increase in headcount focusing on AI-related roles. The shift toward operationalizing AI involves embedding it into business processes and workflows, emphasizing the importance of a secure, governed content layer to ensure accuracy and reliability. As AI becomes a core part of the business landscape, organizations must build comprehensive governance structures to manage and scale these technologies effectively.

Anthropic's Claude Sonnet 5, evaluated using Box's Complex Work Eval benchmark, demonstrates improvements over Sonnet 4.6 in operational domains such as Energy, Retail, Professional Services, and Technology, reflecting its suitability for high-volume enterprise workflows. The benchmark measures models based on multi-step tasks with real business documents, emphasizing final output quality rather than isolated steps. Sonnet 5's streamlined agent loop enhances accuracy in document-heavy operations, reducing reconciliation errors and manual checks, which is crucial for scalable production workflows. This reliability and efficiency make it an attractive option for enterprises transitioning AI from pilot projects to full-scale production, offering consistent performance that supports large-scale deployments. Claude Sonnet 5 will soon be available to Box AI customers, enabling them to leverage its capabilities in their enterprise environments.

Biopharma companies effectively utilizing AI prioritize structuring, classifying, and governing their data before deploying AI capabilities, which prevents costly disorganization and compliance issues. Despite AI's growing use in the industry for data analytics and clinical decision-making, a significant portion of valuable data remains unstructured and inaccessible to traditional AI systems. Companies like a particular Box life sciences customer highlight the importance of establishing a solid content foundation before applying AI, which involves understanding data, classifying it as a trust mechanism, and distributing ownership of content governance among relevant teams. This approach allows organizations to automate metadata extraction, streamline document review processes, and maintain compliance, ultimately enhancing speed, insights, and operational efficiency. By ensuring that AI operates within a governed and auditable framework, biopharma companies can achieve reliable and actionable AI outputs, underscoring the necessity of addressing the content foundation prior to AI model deployment.

The introduction of the new Box npm package streamlines the process of building applications on Box by combining the Box Node SDK and Box CLI into a single installation, making it easier for developers to create full-stack applications without adding a new abstraction layer. This package simplifies development by offering a unified entry point and encourages the use of subpath imports for accessing the SDK, while also providing the CLI for efficient file manipulation and management. An example project demonstrates creating a SpaceX S-1 Q&A chatbot using the Box Agent Skill, Vercel's Chat SDK, and Box's /ai/ask endpoint, highlighting the package's utility in rapidly setting up an app with Box's developer tools. The process involves configuring environment variables, authenticating with Box credentials, and using the CLI to upload necessary files, showcasing the ease and speed of development made possible by this integration. As AI-assisted coding grows in prevalence, the Box npm package represents a shift towards more efficient, bundled developer tools that facilitate quick prototyping and application development.

Valmark, a broker-dealer and investment advisor, revamped its labor-intensive audit process by integrating Box CLI with Claude Code, an AI terminal assistant, turning it into a streamlined, AI-first workflow. This transformation enabled non-technical team members to authenticate easily and execute complex audit workflows through natural language prompts, eliminating the need for developer intervention. By leveraging Box CLI’s help functionality, Claude Code effectively translated natural language into precise API commands, thus simplifying metadata template referencing and reducing manual lookup tasks. The integration also enhanced document processing capabilities, allowing Optical Character Recognition (OCR) of scanned documents and on-the-fly AI extraction of structured data, which improved downstream analysis and made data instantly audit-ready. This approach not only increased efficiency but also empowered Valmark’s Commissions team to manage the audit cycle independently, without extensive scripting knowledge.

Businesses across various industries manage a vast and growing cache of digital content, with effective management ensuring brand consistency, productivity, and risk reduction. However, traditional Digital Asset Management (DAM) systems often fall short due to a lack of metadata and the inability to handle unstructured data efficiently, leading to operational inefficiencies like lost time and duplicated efforts. AI-powered DAM solutions, such as those provided by Box, aim to transform content management by centralizing, securing, and accelerating the content lifecycle with features like AI-driven search, automatic metadata extraction, and automated workflows. These systems enable businesses to better organize their digital assets, streamline processes, and prepare for an AI-integrated future, offering a competitive advantage by making content easier to find, safer to govern, and ready for AI applications. Creative agencies have already seen significant time savings and improved efficiency by using such systems, positioning themselves to leverage AI capabilities effectively in the future.

Enterprise knowledge bases are becoming essential infrastructure for AI agents, requiring a trusted knowledge layer that is shared, permission-aware, versioned, and reviewable. The challenge lies in ensuring that new files are trusted enough for agents to use, as knowledge bases often become disorganized with a mix of current, stale, and never-reviewed content. To address this, a robust intake and review process is necessary, where new knowledge is uploaded, reviewed, and approved before being placed in a curated knowledge base. Tools like Box Automate and Box Hubs help manage this process by routing files for review, assigning tasks, and ensuring only approved content is accessible to agents. This structured approach allows teams to maintain a healthy knowledge base, facilitating dependable AI workflows by providing agents with access to approved and governed content. Maintaining the quality of this context layer is crucial, as companies that manage this effectively will transform everyday work into reliable knowledge for AI integration.

This developer tutorial provides a comprehensive guide on transforming the Hermes AI agent from a personal productivity tool into a collaborative team assistant by integrating it with the Box platform, a shared workspace utilized by teams. The tutorial emphasizes overcoming the common issue of AI agents working in isolation, which can lead to disjointed workflows, by establishing a unified "company brain" that aligns both human and AI collaborators with the same source of truth. It outlines the process of setting up a Box CCG app for Hermes, ensuring that it operates under a dedicated service account to maintain a clean security boundary, while using tools like the Box CLI and Slack to streamline communication and content management. By leveraging these integrations, teams can enhance collaboration across various domains, such as marketing, product launches, engineering, and legal work, enabling Hermes to efficiently manage tasks from a single, updated source of information.

The walkthrough describes the development of a loan origination web application using Next.js and Box to manage document-heavy processes, aiming to simplify the borrower experience. The application integrates Box as a secure content layer for uploading, viewing, and classifying required documents using Box AI and completing credit authorization via Box Sign. Developers are guided through setting up a Box developer account, creating a Box platform app, and configuring the application with specific Box services to streamline workflows and ensure secure document handling. The process involves using downscoped tokens for secure browser interactions, maintaining privileged operations on the server, and leveraging Box's capabilities for document classification and signature capture. Additionally, the guide suggests potential enhancements such as using a real database, adding authentication, and expanding document classification to structured data extraction, setting the stage for scalable, production-grade document workflows in various industries.

Box's 4th annual Family Day, organized by the Families at Box (FAB) Employee Resource Community, has become a significant summer event for the company, engaging 224 employees and 476 of their loved ones with a space-themed day filled with activities like a Mission to Mars Obstacle Course and a virtual puppet show. The event not only celebrated family connections but also emphasized cultural celebrations, such as Juneteenth and Pride Month, with treats and activities sponsored by various Employee Resource Communities (ERCs), including the Black Excellence Network and the Pride ERC. A noteworthy part of the event was the Book Drive, in collaboration with Box.org, which encouraged book swapping among employees and donated surplus books to local nonprofits. Additionally, a fundraiser was launched for the Box Stands Together Fund to support employees facing hardships, with Box.org matching donations 2:1 up to $2,500. This year's Family Day highlighted the company's commitment to community and intersectionality, showcasing the efforts of volunteers who contributed to its success.

The Box HTML Editor integrates HTML file management directly into the Box platform, allowing teams to create, edit, render, and collaborate on HTML files without the need for downloads or external tools, thereby maintaining version control and IT visibility. This native integration ensures that all HTML files inherit Box's enterprise-grade security, automatic version history, and access permissions, making them fully accessible to Box AI for querying and layout generation. The editor offers three distinct viewing modes—code view, preview mode, and split view—catering to various user needs, from developers to marketers. By centralizing HTML editing within Box, organizations can eliminate tool friction, preserve governance, and facilitate AI-ready content creation, ultimately embedding HTML files into the organization's governed content layer for secure, collaborative, and AI-driven workflows.

Law firms are grappling with the challenge of integrating AI into their operations due to reliance on outdated, fragmented content management systems that hinder efficiency and governance. These firms produce vast amounts of documentation essential to their operations, but the lack of a centralized, governed content infrastructure results in inefficiencies and missed opportunities for AI integration. Bonnie Kennedy, Director of Information Governance at Fisher Phillips, illustrates this challenge as she transitions from manual processes to a more automated, governed system using Box. The key to unlocking AI's potential in law firms lies in establishing a robust content foundation that centralizes, governs, and activates documents, ensuring that AI can operate effectively and securely. This approach not only streamlines workflows and reduces manual inefficiencies but also enhances client relationships by providing a more seamless and reliable service.

The narrative of AI replacing jobs is being challenged by emerging data that suggests AI is actually creating new roles and transforming existing ones across industries. According to the Box 2026 State of AI in the Enterprise report, 58% of surveyed IT decision-makers anticipate an increase in their organization's headcount over the next three years, with AI's integration leading to the development of jobs such as agent operators, governance professionals, and workflow specialists. AI is not merely automating tasks but is also enhancing human work, particularly in sectors like healthcare, manufacturing, and financial services, where it aids in quality control, predictive maintenance, and fraud detection. The demand for roles that require analytical, technical, or creative skills has increased, while jobs involving repetitive tasks have decreased. Human-AI collaboration is reshaping the labor market, necessitating roles that bridge business knowledge with AI oversight, such as AI governance and ethics specialists, AI solutions architects, and workflow automation specialists. Organizations that have built a robust content infrastructure are leveraging AI most effectively, creating a competitive advantage and enabling new roles that focus on managing and optimizing AI-driven workflows. This shift reflects AI's role as a capability expander rather than a replacement for human effort, allowing workers to focus on more strategic and satisfying tasks.

Enterprise CIOs face the challenge of integrating AI agents effectively within organizations by establishing a shared, permission-aware, and curated enterprise knowledge base that prevents operational risks from inconsistent information across different agents. As AI agents become more prevalent, it's crucial for organizations to avoid creating silos by ensuring these agents access a centralized knowledge layer that is versioned, reviewable, and accessible through APIs and various integrations. Domain-specific knowledge bases should be maintained by relevant teams to ensure the highest quality and accountability, allowing agents to access tailored content for specific tasks. This approach transforms the knowledge base into a reusable enterprise capability, enhancing the consistency and reliability of AI-generated work and fostering an organizational memory that continuously improves through feedback from human and agent workflows. For CIOs, the goal is to build this knowledge infrastructure to ensure AI agents operate consistently, thereby giving enterprises a competitive advantage in the evolving AI landscape.

Box Doc Gen and Box Sign streamline document workflows by automating the generation and e-signature processes. This guide demonstrates the integration of these APIs to automate offer letter creation and delivery, using a sample app triggered by an applicant tracking system such as Greenhouse or Workday Recruiting. The app merges candidate data into an offer letter template with Box Doc Gen, and upon PDF generation, triggers a Box Sign request for e-signature. The setup requires a Box Enterprise Advanced account and involves configuring app scopes, enabling webhooks, and setting up a server using Node.js. The solution allows for customization, including multiple signers and approvers, and supports various workflows across sectors like HR, sales, legal, and finance by facilitating the seamless creation and approval of documents requiring signatures.

The evolution of web software from relational databases to modern AI-driven interfaces has led to a shift in how agents process information, with a focus on developing effective read/write memory systems. Traditional interfaces like SQL and APIs, while designed for deterministic software, present challenges for agents that must navigate complex, dynamic environments. Instead, leveraging the foundational knowledge of filesystems, which large language models (LLMs) have been extensively trained on, presents a promising solution for agent memory architectures. Filesystems offer a familiar, inspectable, and revisable structure that aligns well with agents' operational needs, allowing for seamless interaction between human and machine. This approach highlights the potential of using systems like markdown files for storing and managing data, emphasizing shared legibility and collaboration. By aligning agent interfaces with paradigms that models inherently understand, such as filesystems, email, or spreadsheets, developers can streamline agent operations, reduce complexity, and enhance debugging capabilities, all while meeting enterprise requirements for collaboration and auditability.

The text explores the enduring significance of records and files throughout history, arguing that every technological advancement—from Mesopotamian clay tablets to modern AI—has emphasized the need for stable, accountable records rather than replaced them. As the volume of data grows and AI agents become more prevalent, the need for reliable systems of record with properties such as identity, provenance, access control, versioning, and structure intensifies. The author contends that although AI agents are powerful readers of information, they depend on robust content platforms that ensure records are legible and verifiable by humans, especially when errors occur. The text highlights that, while models and agents evolve rapidly, the underlying accountability layer has been underdeveloped, and it will be crucial for future AI systems. This layer enables enterprises to stand by their data-driven decisions and aligns with historical lessons that emphasize the importance of preserving knowledge in a human-readable form.

Anthropic's release of Claude Fable 5 marks a significant development in AI cybersecurity capabilities by making the first Mythos-class model accessible to the public, while reserving the full capabilities of Mythos 5 for select government and critical infrastructure operators in the Project Glasswing initiative. This release includes protective safeguards that redirect certain queries to a previous model, Opus 4.8, to prevent misuse, indicating a cautious approach as Anthropic fine-tunes its system. Although Fable 5 shows potential in identifying vulnerabilities, its effectiveness is moderated by these safeguards, which can impact its use in security programs. The release also introduces a 30-day data retention policy, sparking discussions on data governance and security implications. The challenge now lies in preparing security teams for the increased volume of vulnerability discoveries that AI models like Mythos can generate, emphasizing the need for enhanced remediation processes and automation to manage and act on these findings effectively.

Utah State University has transformed its document management and workflow processes by implementing Box Extract, an AI-powered tool that automates data extraction with a 99.7% accuracy rate across over 2,500 financial documents. Originally adopted as a cloud storage solution, Box now serves as the university's primary platform for collaboration and file management, replacing systems like OneDrive. The integration of Box with ServiceNow has further streamlined operations, allowing for seamless workflow automation and efficient handling of unstructured content. This transformation has not only improved financial documentation and compliance but also facilitated various academic and administrative functions, such as policy management and student-professor collaboration. By leveraging innovative technologies, Utah State University has set a precedent for how higher education institutions can effectively modernize their processes to meet compliance requirements and enhance operational efficiency.

In an exploration of the evolving challenges in security operations, the text discusses the complexities introduced by interconnected AI agents that operate across multiple systems, emphasizing the risks when these agents fabricate conclusions or reach unwarranted verdicts due to insufficient data. Traditional security models, which focus on defined roles and predictable interactions, fall short in managing the dynamic and autonomous nature of AI agents that can inadvertently expose sensitive information through their interconnected workflows. The text argues for a shift from access control to execution control, advocating for security measures that govern the behavior and coordination of agents rather than just their initial access permissions. It highlights the necessity for a more governed, policy-aware approach to data, suggesting that companies that manage this effectively will have more predictable AI agent behavior. The discussion underscores the importance of understanding system behavior over time and ensuring that actions align with intended outcomes, as AI agents increasingly interact with each other rather than with humans directly.

An enterprise AI strategy is a comprehensive framework designed to align AI initiatives with specific business objectives, ensuring scalable and impactful implementation across organizations. The strategy emphasizes the importance of establishing a strong data foundation, governance, and architecture before deploying AI technologies to avoid the common pitfalls that lead to failure in enterprise AI programs. These pitfalls often include fragmented data systems, misalignment between business and technical teams, and pilots that cannot transition into production due to inadequate infrastructure and governance. Successful enterprise AI strategies require a coherent plan that links business priorities with data readiness, risk management, architecture, and workforce adoption. This approach allows AI systems to integrate seamlessly into existing business operations, delivering measurable outcomes such as improved efficiency, reduced costs, and enhanced customer experiences. Additionally, the strategy must encompass content governance, ensuring AI can access and utilize trusted enterprise information, and must include change management plans that prepare the workforce for AI adoption. Organizations are encouraged to build, buy, or partner based on their maturity level in data and machine learning capabilities, and to focus on KPI-linked use cases for a tangible return on investment.

Founders are increasingly creating personal AI knowledge bases that organize and utilize scattered information such as notes, research, and strategy documents, using agents like Claude Code or Codex. While effective for individual use, these personal vaults fall short when AI becomes a team workflow, necessitating a shared source of truth for teams. This evolution has led to the concept of a "shared company brain," a governed repository that integrates company context, product knowledge, customer information, operating rules, and generated outputs, enabling multiple users and agents to access, update, and generate outputs from a common, trusted context. Governance is crucial as it provides permissions, version control, and audit trails, ensuring that the shared knowledge base maintains integrity and trustworthiness across diverse team members and tools. Box emerges as a solution by serving as the governed content layer, allowing both local and cloud-based agent workflows to operate from a single source of truth, thus optimizing team alignment and enhancing operational efficiency in AI-native companies.

Box's State of AI in the Enterprise report 2026 explores the impact of AI on the workforce, challenging the common prediction that AI will drastically reduce jobs. Contrary to reports from Goldman Sachs and the IMF, which warn of significant job displacement, the report reveals that many organizations, particularly those advanced in AI adoption, anticipate workforce growth. While only a small percentage currently see significant job elimination due to AI, many organizations are shifting towards a "digital workforce strategy," where AI agents take on defined roles, allowing humans to focus on oversight and exception handling. This shift has led to the creation of new roles, such as AI agent operators and workflow automation specialists, demonstrating a pattern where AI-driven efficiencies lead to increased demand and job creation. The report suggests that as AI reduces the cost of work, demand expands, resulting in a workforce that, while differently structured, is larger and centered around new capabilities.

The "State of AI in the Enterprise 2026" report highlights that while a majority of organizations have adopted AI agents and report measurable returns on investment, the most significant benefits are concentrated among those at the leading edge of AI maturity. These leading organizations have successfully connected AI agents to trusted content, established robust governance frameworks, and prioritized flexibility, showcasing a distinct operational maturity. The report indicates that while belief in AI's potential is widespread, the key differentiator is the implementation and integration of AI capabilities, with leading-edge organizations demonstrating advanced usage and visibility into AI applications compared to those at earlier stages. The findings suggest that the gap between belief in AI's potential and actual operational execution is significant, with the leading edge setting the benchmark for effectively leveraging AI in enterprise settings.

AI adoption has rapidly expanded across enterprises, with 99% of organizations utilizing AI in some form, and a significant impact is evident as 80% report moderate or significant ROI from AI implementations. The report highlights a maturity gap, with leading-edge companies achieving greater returns by using AI for innovative purposes rather than just efficiency improvements. These advanced organizations are not only automating tasks but are also engaging in entirely new types of work, seeing broader impacts on productivity, workflow innovation, and new revenue streams. A distinct operational model for AI is emerging, with a "managed digital workforce" approach, where AI agents are integrated and managed like employees, supported by a structured AI leadership role in 95% of organizations. A hybrid economic model for AI costs, where IT manages infrastructure while business units cover application-level expenses, is becoming prevalent, facilitating a convergence between IT and business objectives. Despite structured management and incentives for AI use, the most successful AI initiatives often arise from collaboration across business units and IT functions, rather than solely from centralized AI or data-science teams.

In the 2026 State of AI in the Enterprise report by Box, the primary challenge identified is not the capability of AI models but the accessibility and trustworthiness of enterprise knowledge for AI systems. While 96% of organizations acknowledge the importance of AI agents accessing company-specific content, only 36% have effectively connected these agents to reliable internal content across various use cases. The gap is most pronounced at the extremes of organizational maturity, with leading-edge companies more successfully integrating AI with unstructured data as a competitive advantage. The report highlights that operational barriers, such as fragmented data and legacy systems, pose significant challenges to effective AI deployment by hindering agents' ability to interact with organizational content. As AI evolves, the focus shifts from model access to content access, emphasizing the necessity of quality information and robust governance to ensure AI systems function effectively within enterprises.

Chapter 5 of Box's "State of AI in the Enterprise" report for 2026 highlights the increasing trend of organizations adopting a multi-model approach to artificial intelligence, with an average of 3.3 AI tools used by respondents. This shift is driven by the desire to remain adaptable and avoid lock-in with a single provider, as 68% express concern about dependency on one model. The concept of headless operation for AI agents, which enables them to function directly across systems and workflows without a human interface, is gaining popularity, with 80% deeming it important or critical. The adoption of these practices is more prevalent among organizations at the leading edge of AI maturity, demonstrating the evolving landscape where flexibility and scalability are prioritized.

Box's 2026 State of AI in the Enterprise report, developed from a survey of 1,640 IT decision-makers across the US, UK, France, and Japan, highlights a dramatic shift toward AI integration with a rise in organizations identifying as advanced or leading-edge from 8% to 64% in just a year. This evolution is marked by significant ROI, with 80% reporting at least a 10% improvement, and leading-edge companies achieving over 25% ROI by operationalizing AI through multi-step workflows, governance, and flexible infrastructure. Key trends include using AI for new work rather than just cost reduction, the expectation of workforce growth due to new AI roles, and the need for AI governance and platform flexibility to avoid vendor lock-in. The report underscores the importance of integrating AI agents with company-specific content and emphasizes that leading-edge companies excel by building the infrastructure for an agentic enterprise, allowing them to adapt as the AI landscape evolves.

Claude Fable 5, developed by Anthropic, represents a substantial upgrade in enterprise document intelligence, achieving 83% accuracy on Box's Complex Work Eval benchmark, compared to 79% for its predecessor, Opus 4.8. The improvement is particularly notable in tasks requiring report drafting from data, financial analysis, and multi-step document reasoning, where Fable 5 demonstrates superior numerical precision and reasoning about edge cases. This model's advancements are most pronounced in content-heavy and numerically complex industries such as Media & Entertainment, Technology, and Financial Services, with notable accuracy improvements over Opus 4.8. Fable 5 displays enhanced consistency and predictability across repeated runs, making it a more deployable solution for enterprise workflows that demand precision and reliability. As a result, Fable 5 is poised to offer significant operational advantages to Box AI customers, with its availability as a Beta model in Box AI Studio.

The 2026 State of AI in the Enterprise report highlights the critical role of governance in the effective scaling of distributed AI within organizations, emphasizing that governance designed specifically for AI agents, rather than retrofitted from human workflows, can provide the necessary visibility and control. The report reveals that while stronger governance might seem to slow adoption, it actually empowers organizations to manage incidents better, with leading-edge companies reporting more AI-related incidents due to their enhanced capacity for detection and visibility. Despite a significant increase in organizations establishing AI governance frameworks, many still lack comprehensive visibility and formal standards, indicating that the real challenge lies not in AI model capability but in organizational capability. As companies progress in AI maturity, they integrate governance with infrastructure, permissions management, and knowledge access into a cohesive operational model, reflecting a shift from ad hoc measures to more robust frameworks.

AI offers significant opportunities for enterprises to leverage vast amounts of content, yet also presents novel security challenges, particularly concerning unstructured data governance. The discussion between Heather Ceylan, Box's Chief Information Security Officer, and Dave Bittner from CyberWire highlights the critical issue of managing unstructured data that is dispersed across various platforms, potentially leading to unauthorized access when AI agents interact with it. Traditionally, this has been less of a concern with human access due to limited scope; however, AI's ability to access and act on extensive datasets increases the risk of data breaches. To mitigate this, organizations are advised to classify and label data, applying stringent permissions to control access. AI can assist in this classification process, making it more feasible than manual efforts, thus enhancing security and productivity while maintaining necessary human oversight for sensitive operations. This approach aims to harmonize AI utilization with robust security measures, allowing enterprises to benefit from AI while safeguarding their data.

Box transitioned its AI agent platform to a multi-step agentic system using LangGraph as its core execution engine, opting against building a custom solution from scratch. This decision enabled the development of the Box Agent platform, which is structured into five distinct layers: entry points, an intelligence service, an Agent Orchestrator, an LLM gateway, and model providers. LangGraph's graph-based execution model supports branching, parallel execution, and real-time event streaming, allowing for dynamic task planning and seamless execution at an enterprise scale. Agents, defined using Box’s Agent Definition Language, are compiled into Deep Agents at runtime, facilitating behavior updates without code deployments. The platform's architecture supports resumability, enabling interrupted sessions to be resumed from the latest checkpoint. The system's modularity and scalability allow it to handle both simple and complex workflows, with the Agent Orchestrator managing the full lifecycle of agent execution across multiple teams at Box. The integration with LangGraph and the use of Deep Agents provide a robust framework for handling dynamic tasks and live configuration changes, ensuring reliability and developer velocity without sacrificing the complexity of managing multiple provider integrations.

Sales workflows often extend beyond CRM data, requiring the creation and management of documents like business plans or approval packets. By integrating Box Doc Gen and Box Automate via API, a seamless process can be established where CRM opportunity data generates an approval document, routes it for review, and updates Salesforce automatically. This setup involves configuring a Box Platform app to work with Box Doc Gen for document creation and Box Automate for workflow management, which includes sending documents for approval and updating CRM records. Key steps include setting up webhooks, creating metadata templates, and configuring the Box Automate workflow to handle document approval tasks and CRM updates. The integration allows for automating document approval processes and ensuring the final status is updated in the CRM, enhancing efficiency across industries.

Box, a company that has grown from a startup to a global enterprise, faced increasing complexity in its IT environment as systems became highly interconnected and dependencies harder to manage. To address this, Box established the IT Domain Architecture Group (ITDAG) to enhance architectural alignment across its IT domain, aiming to improve decision-making and risk management. ITDAG is a collaborative forum designed to identify cross-team dependencies, integrate security, legal, and compliance considerations earlier, and foster a culture of shared architectural responsibility. By engaging stakeholders from various IT functions early in the design phase, ITDAG helps prevent operational challenges and ensures that architectural decisions are well-informed and aligned with the company's broader IT ecosystem. Through this approach, Box has succeeded in creating a scalable framework that emphasizes ongoing collaboration, operational resilience, and long-term maintainability, ultimately fostering a shared architectural mindset within the organization.

In the evolving landscape of enterprise AI, the focus is shifting from purely model-centric approaches to workflow-oriented solutions, placing emphasis on orchestrating and managing content-driven processes. While AI models have demonstrated significant capabilities, the true value in enterprise settings lies in integrating these models within robust workflows that address the intricacies of organizational operations, such as document handling, multi-party reviews, and compliance with changing policies. Box Automate exemplifies this transition by embedding AI-powered automation directly within content management systems, ensuring that essential elements like permissions, metadata, and audit trails are seamlessly maintained throughout the process. This approach not only enhances reliability and reduces operational ambiguities but also enables enterprises to leverage AI in a manner that aligns with their existing workflows and regulatory requirements. By prioritizing governance and context preservation, Box aims to redefine enterprise AI as a tool for enhancing workflow efficiency, rather than as an isolated technological marvel.

Banking onboarding workflows face significant challenges due to their reliance on fragmented systems and manual processes, leading to high abandonment rates and dissatisfied customers. With up to 68% of consumers dropping out of online applications, the friction in traditional processes is evident, especially in wealth management and commercial banking, where complexity increases due to additional requirements like trust structures and multi-party KYC reviews. The solution lies in consolidating these disparate elements into a single, governed, end-to-end workflow that integrates document collection, AI-assisted review, approvals, and signatures, thereby enhancing efficiency and transparency. For example, Mercer Advisors significantly improved their client onboarding process by reducing quote turnaround times from two weeks to five minutes and achieving substantial productivity gains by automating document processing. Ultimately, modernizing onboarding workflows is crucial for banks to meet client expectations of a seamless digital experience, streamline operations, and build lasting trust from the outset of the client relationship.

Denver's approach to integrating AI into its public services, led by CIO Suma Nallapati, emphasizes practical applications that solve real organizational problems, particularly in aiding vulnerable populations. The city introduced Sunny, an AI-driven virtual assistant, to enhance customer service by providing residents with instant access to information and services in multiple languages, significantly reducing operational costs. This initiative illustrates a broader strategy where AI is used not to replace human workers but to alleviate routine tasks, allowing public servants to focus on more meaningful work. Nallapati stresses the importance of a secure, well-organized content environment for effective AI deployment, underscoring data governance and ethical considerations. By automating repetitive processes, Denver has achieved significant efficiency gains while maintaining service standards despite budget constraints. The city's experience highlights the value of AI in amplifying human expertise through better content curation, governance, and human oversight, fostering trust and enabling faster access to crucial information without compromising data security.

The text explores the challenges and solutions related to enabling AI agents to manage binary files in enterprise environments without involving the LLM context window, focusing on issues such as context bloat, payload limits, latency, and data corruption associated with base64 encoding. By implementing a system that uses temporary, single-use signed URLs for direct file transfers between Box storage and local execution environments, the architecture bypasses these problems, ensuring secure, compliant, and efficient file handling. This approach not only preserves the integrity and security of the data but also maintains comprehensive audit trails through OAuth session binding and admin opt-in controls, allowing enterprises to exercise granular control over this capability. The integration of signed URL tools into the Box MCP server enables AI agents to perform complex workflows like editing and re-uploading documents at scale, closing the gap between static and live files while maintaining robust enterprise governance.

Box has introduced Search 3.0, a significant AI-powered upgrade to its search infrastructure, designed to address the longstanding challenges of enterprise search and enhance the speed, relevance, and intelligence of content discovery. This new system drastically reduces search latency and file indexing time from 15 minutes to mere seconds, improving both Quick and Full Search functionalities. The upgrade allows for near-real-time indexing and supports external and native AI agents in performing deep-document searches via Box APIs, facilitating more accurate and contextually relevant search results. The enhanced Box Search is poised to transform enterprise search by accommodating multimodal content, hybrid search, and natural language retrieval, while offering developers powerful public APIs for building advanced search experiences. This development positions Box to better meet the needs of information-driven organizations by unlocking the intelligence within their content for users, teams, and AI agents.

Box CEO Aaron Levie emphasizes the challenges companies face in integrating AI into business processes, despite the technology's rapid advancements and capabilities. While many businesses achieve quick wins like enhanced productivity through chatbots, the real challenge lies in effectively embedding AI into core operations. This is where Forward Deployed Engineers (FDEs) come in, tasked with bridging the gap between AI's potential and its practical application within enterprise contexts. Box AI architects Alex Leutenegger and Gilbert Ortega-Rivera highlight the importance of 'context engineering,' which involves tailoring AI solutions to fit specific business needs and processes rather than merely adjusting AI models themselves. They illustrate how understanding and refining the enterprise's unique data and context can drastically improve AI performance, as seen in examples like medical chart classifications and offer evaluations for touring musicians. The role of FDEs is crucial as they combine expertise in business-process transformation and proficiency with large language models (LLMs) to continually adapt and improve AI systems in response to evolving technological capabilities.

Enterprise AI projects often falter not due to insufficient model power but because they lack integration with the business context, such as contracts, records, and workflows that hold organizational knowledge. Box Forward Deployed Engineers (FDEs) address this challenge by preparing business content for AI, designing AI-centric workflows, and continuously improving these systems as models and approaches evolve. These roles blend engineering, solution architecture, and customer interaction, ensuring AI solutions are aligned with specific business needs and industry contexts. FDEs work closely with businesses to transform high-value processes, selecting the best models for specific tasks and adapting strategies over time. Engagements with an FDE can vary from short-term assessments to ongoing collaborations, allowing organizations to continually refine their AI implementations and achieve significant business improvements.

Agentic AI governance is the structured management of autonomous AI systems that execute tasks on behalf of organizations, emphasizing authority control rather than just output quality. This approach is essential for roles like chief AI officers and CISOs who manage workflows involving content and data, as it addresses risks such as unauthorized actions, data exfiltration, and privilege escalation. Unlike traditional AI governance, which focuses on the quality of model outputs, agentic AI governance ensures that actions are authorized and within defined boundaries. The guide provides an eight-step framework for implementing agentic governance, highlighting the importance of defining the agent's scope, maintaining strict identity and access boundaries, and establishing human oversight thresholds. It also discusses how Box manages agentic AI governance at the content layer by ensuring that agents operate within the same security and compliance boundaries as human users. The document stresses the need for continuous monitoring to prevent authority expansion and outlines the responsibilities of various stakeholders, including model providers, platform operators, integrators, and deploying organizations, to ensure accountability when autonomous agents are deployed.