Box CEO Aaron Levie on AI’s ‘era of context’

During the recent Boxworks developer conference, Box made waves by unveiling a suite of innovative AI capabilities, embedding advanced AI models into its core product offerings. This year’s conference saw a notable increase in product announcements, reflecting Box's rapid advancements in AI technology. Following the launch of its AI studio last year and the introduction of data-extraction agents earlier this year, Box is now presenting Box Automate, a transformative system designed to optimize workflows by integrating AI agents intelligently. In an exclusive conversation with CEO Aaron Levie, the discussion centered on Box's strategic approach to AI integration and the challenges of competing with major players in the foundation model landscape. Levie expressed enthusiasm about the potential of AI agents to reshape the modern workplace while maintaining a realistic perspective on the current limitations of these technologies. The core of Box's vision is to address the evolving nature of work driven by AI, particularly in the realm of unstructured data. Levie highlighted that while significant strides have been made in automating structured data processes through systems like CRM and ERP, unstructured data remains largely untapped. He emphasized the complexities involved in workflows such as legal reviews and marketing asset management, which require human intervention due to the challenges of document comprehension by computers. The conversation also touched on the anxieties businesses face regarding the deployment of AI agents, especially in sensitive contexts. Levie noted that clients seek assurance that AI workflows will operate consistently and predictably, avoiding errors that could escalate over time. Box Automate allows companies to define how much responsibility each AI agent has within a workflow, ensuring a balance between automation and oversight. Moreover, Levie addressed the limitations of existing AI models, particularly regarding context management. He explained that as tasks grow in complexity, AI models can struggle to maintain effective decision-making. By segmenting workflows and utilizing sub-agents, Box aims to provide the necessary context to AI agents, enhancing their effectiveness and reliability. The discussion also explored the ongoing industry debate surrounding the merits of large, powerful AI models versus smaller, more reliable alternatives. Levie clarified that Box's system is designed to accommodate various task complexities without compromising on control and governance. He emphasized the importance of creating a robust architecture that adapts alongside evolving AI capabilities. Data security remains a critical concern, as Levie pointed out the risks of granting AI unrestricted access to sensitive information. He underscored the necessity for stringent access controls and compliance measures to prevent unauthorized data exposure. Box's extensive experience in data governance positions it to tackle these challenges effectively, ensuring that AI agents operate within defined boundaries. Finally, the conversation shifted towards the competitive landscape, particularly the emergence of new functionalities from foundation model companies. Levie reiterated that enterprises require a comprehensive solution that prioritizes security, control, and flexibility in AI deployment. Box's commitment to providing a secure, adaptable platform positions it uniquely in the market, capable of integrating with various AI models while maintaining essential functions like storage and permissions management.