The rise of AI agents is changing how enterprises handle unstructured data, and Box is positioning itself at the center of this transformation. At BoxWorks 2025, CEO Aaron Levie announced Box Automate, a powerful new system designed to integrate AI into everyday workflows.
Levie describes this as the “era of context” in AI, where businesses must unlock value hidden in documents, contracts, and creative assets—not just structured databases. While the potential is massive, Levie acknowledges that AI is still bound by limitations, requiring careful guardrails to ensure reliability, security, and compliance.
In this post, we’ll break down Levie’s vision, the new Box Automate system, and what it means for the future of AI-driven business processes.
Box’s AI Journey: From Studio to Automate
Over the past year, Box has rapidly expanded its AI capabilities:
- Box AI Studio launched last year, allowing developers to experiment with agentic AI models.
- Data-extraction agents were introduced in February 2025.
- Search and deep research agents followed in May.
Now, Box is rolling out Box Automate, which Levie describes as an “operating system for AI agents.” This platform breaks workflows into smaller, manageable segments where AI can be deployed efficiently and securely.
Why Box Is Betting on Unstructured Data
Automation in structured data (CRM, ERP, HR systems) is nothing new. But Levie points out that most enterprise workflows revolve around unstructured data—documents, contracts, creative files, legal reviews, and more.
“For the first time ever, AI agents can actually tap into all of this unstructured data,” Levie explains.
Some key use cases include:
- Legal document review for contracts and compliance.
- Marketing asset management, such as campaigns and digital assets.
- M&A deal workflows, where teams must analyze thousands of sensitive files.
By focusing on unstructured data, Box hopes to close a critical gap in enterprise AI adoption.
Box Automate: Guardrails for AI at Scale
One of the biggest risks in deploying AI agents is unpredictability. Enterprises can’t afford “runaway” models that compound mistakes. Box Automate solves this by:
- Breaking workflows into segments – for example, separating a submission agent from a review agent.
- Adding deterministic guardrails – ensuring agents act consistently and don’t drift off-task.
- Balancing deterministic and agentic AI – businesses can choose where they want precision and where they allow flexibility.
This architecture ensures scalability, reliability, and compliance, making AI safer to deploy in sensitive industries.
The “Era of Context” in AI
Levie describes the current stage of AI development as the era of context.
While foundation models like Claude or GPT-4 are powerful, their context windows are limited. This means they can’t always process long or complex tasks effectively. By breaking workflows into smaller tasks, Box ensures that AI models can operate within their context limits without losing accuracy.
In Levie’s words:
“There’s no free lunch right now in AI. What AI models and agents need is context, and that context comes from unstructured data.”
Security, Permissions, and Trust: Box’s Advantage
One of the top enterprise concerns with AI is data security. Foundation models risk regurgitating sensitive data, but Box’s strength lies in decades of building secure enterprise storage and access control systems.
With Box Automate, organizations can:
- Enforce strict data permissions and governance.
- Ensure agents only access information a user is authorized to see.
- Maintain compliance across industries like healthcare, finance, and law.
This security-first approach could give Box a strategic advantage over foundation model competitors.
Competing with Foundation Model Companies
AI-first companies like Anthropic and OpenAI are rapidly moving into enterprise solutions. Anthropic’s recent Claude.ai file upload feature is one example.
But Levie believes Box has an edge:
- Security & governance baked in.
- Choice of AI models so companies aren’t locked into one vendor.
- Powerful APIs and integrations for enterprise-scale deployment.
- Vector embedding + storage built natively into the platform.
In short, Box isn’t just offering AI—it’s offering future-proof enterprise AI infrastructure.
Key Takeaways from Aaron Levie’s Vision
- Unstructured data is the next big frontier in enterprise AI.
- Box Automate provides a secure, flexible “operating system” for deploying AI agents.
- We’re in the era of context, where workflows must be segmented to overcome model limitations.
- Security, compliance, and permissions remain the core differentiators for enterprise AI success.
FAQs About Box AI and Enterprise Workflows
1. What is Box Automate?
Box Automate is Box’s new AI-driven system that acts like an operating system for enterprise workflows, enabling companies to deploy AI agents securely and at scale.
2. Why is unstructured data important in AI?
Most enterprise work involves unstructured data—documents, legal files, marketing assets—which has historically been difficult to automate. AI agents now make it possible.
3. How does Box ensure AI security?
Box leverages decades of experience in enterprise storage and access controls, ensuring that AI agents can’t access unauthorized or sensitive data.
4. How is Box different from foundation model companies?
Unlike standalone AI models, Box provides a secure, governed, multi-model platform that integrates directly with enterprise workflows.
5. What does “era of context” mean in AI?
It refers to the idea that AI performance depends on giving models the right context—breaking tasks into manageable segments to avoid context window limitations.
