Box is betting big on AI’s next chapter.
At its annual Boxworks conference on Thursday, the cloud content company unveiled Box Automate, a new “operating system for AI agents” designed to tackle the hardest problem in enterprise: making sense of unstructured data.
CEO Aaron Levie calls it the “era of context,” where agents finally bring order to messy legal docs, marketing assets, and contracts. But he’s also blunt about AI’s limits — and the guardrails needed to keep agents from going rogue.
Key Takeaways
- Box launches Automate, an AI agent system for enterprise workflows.
- CEO Aaron Levie frames this as AI’s “era of context.”
- Focus is on unstructured data like contracts, assets, and legal docs.
- Guardrails and sub-agents address risks of runaway AI tasks.
- Box positions itself as a secure, multi-model platform for enterprises.
Box CEO Aaron Levie says we’ve entered the “era of context,” where AI agents can finally automate unstructured data workflows like contracts and marketing assets. At Boxworks, the company launched Box Automate, a system that breaks processes into agent tasks while ensuring security, compliance, and enterprise-grade control.
Box’s Big AI Bet at Boxworks
Box kicked off its developer conference, Boxworks 2025, with one of its most ambitious product announcements to date: Box Automate, a framework for deploying AI agents at scale across enterprise workflows.
The move underscores the speed of Box’s pivot into AI. After launching its AI studio last year and adding data-extraction and deep-research agents in February and May, Box is now positioning itself as the go-to platform for automating business processes that rely on unstructured data — from contracts and compliance reviews to M&A deal rooms.
“For the first time ever, we can actually tap into all of this unstructured data,” Levie told TechCrunch in an interview.
Why Box Sees the ‘Era of Context’
Most enterprise automation so far has focused on structured data — the kind of rows and columns that live in ERP or CRM systems. But the majority of business knowledge lives in documents, images, and communications.
Levie argues this is where AI can finally change the game. He describes the industry as entering an “era of context”: giving AI models the relevant, permissioned context they need to act intelligently.
Without that, he warns, “there’s no free lunch.” Large models eventually run out of context window, making long-running autonomous agents prone to failure.
Guardrails Against Runaway Agents
One of the key innovations in Box Automate is the ability to split workflows into multiple specialized agents. For instance, a submission agent might handle intake, while a review agent validates the work before passing it along.
This approach is Box’s answer to customer concerns about reliability and safety. Levie emphasized that enterprises want deterministic guardrails — ensuring an agent won’t “run wild” after hundreds of iterations.
Security and Permissions at the Core
If AI’s promise is automation, its risk is exposure. Enterprises handling sensitive documents can’t afford hallucinated access to restricted data.
Box leans on its decades of experience in access control, compliance, and governance. Levie says this security layer means AI agents in Box cannot serve up answers from data a user isn’t authorized to see.
That’s a sharp contrast to the “give everything to the model” approach some startups have taken — often at their peril.
Competing With the Model Builders
Anthropic, OpenAI, and Google are racing to deliver ever-bigger foundation models. Anthropic’s Claude recently added file-upload capabilities, bringing it a step closer to enterprise content management.
But Levie downplays direct competition. “Enterprises need security, permissions, control, UI, APIs, and freedom of model choice,” he said. Box’s strategy is to remain model-agnostic while serving as the secure orchestration layer on top.
That means storage, embeddings, and workflow logic live in Box, while customers can swap in whichever large language model (LLM) best fits their needs.
AI in the Enterprise
Box’s push fits a broader trend: enterprises are moving past experimentation and into deployment. Analysts predict that by 2026, more than half of enterprise AI deployments will involve multi-agent systems designed to handle complex workflows.
The challenge will be balancing ambition with trust and reliability. Early deployments of autonomous agents have revealed both their potential and their brittleness.
Risks Ahead
Even Levie acknowledges the limits. Current AI models are bounded by context windows and still prone to error. Breaking workflows into sub-agents mitigates some risks, but doesn’t eliminate them.
The success of Box Automate may hinge on whether enterprises trust Box to manage these risks better than startups or hyperscale model providers.
What Happens Next
Box Automate is rolling out to enterprise customers in phases, with broader availability expected later this year.
Industry watchers will be tracking:
- Adoption rates among Box’s Fortune 500 clients.
- Competition from native AI platforms like Microsoft Copilot and Google Workspace AI.
- Model-agnostic execution, ensuring customers can mix and match providers.
Conclusion
Box is making a high-stakes bet that the future of AI is not just bigger models but smarter context. With Automate, it wants to become the control plane for enterprise AI agents.
Whether Levie’s “era of context” takes hold will depend on adoption — and how well Box balances innovation with the caution enterprises demand.
Source TechCrunch