Box announced a new suite of artificial intelligence (AI) features, including the 'Box Automate' system, at its annual developer conference, Boxworks. The company highlighted the integration of agentic AI models across its product portfolio, aiming to enhance workflows involving unstructured data.
Box Automate is positioned as an operating system for AI agents, designed to segment workflows and integrate AI capabilities where necessary. This launch marks a further acceleration of Box's AI development efforts, following the introduction of its AI studio last year and new data-extraction agents in February, alongside search and deep research agents in May.
Aaron Levie, CEO of Box, expressed a positive outlook on the potential of AI agents in enterprise environments. He emphasized that AI's most significant impact currently lies in automating workflows that depend on unstructured data, such as legal reviews, marketing asset management, and merger and acquisition (M&A) deal assessments. Levie noted that while structured data workflows have seen years of automation through systems like CRM and ERP, unstructured data has largely remained untouched by similar advancements until now.
Levie acknowledged the inherent limitations of current AI models, particularly concerning the context window in long-running tasks. Box Automate addresses this by enabling the deployment of sub-agents and establishing 'deterministic guardrails' within workflows. This approach allows organizations to define how much work each agent performs before handing off tasks, thereby managing complexity and maintaining reliability.
Regarding data security and control, Box leverages its existing infrastructure for access controls, data security, permissions, governance, and compliance. Levie stated that this foundational system ensures AI agents can only access data to which a user has explicit permission, mitigating concerns about sensitive information misuse. The company aims to provide a 'future-proof architecture' that supports a range of leading AI models, offering enterprises flexibility and avoiding vendor lock-in, according to Levie.
