AI BriefingHow Box AI Built Enterprise Content Agents with Deep Agents
Key point
Box built an agent system that performs complex enterprise data analysis by leveraging the Deep Agents architecture.
Details
Box developed Box Agent, which goes beyond simple document Q&A to analyze vast amounts of unstructured enterprise data and generate reports. To achieve this, instead of the existing fixed sub-agent approach, Box adopted a flexible architecture based on Deep Agents.
There are two core reasons why Box chose Deep Agents.
- Model Agnosticism: Securing the flexibility to freely choose among various LLM providers such as OpenAI, Anthropic, and Google
- Speed of Iteration: A 3x improvement in development speed by focusing on solving enterprise-specific problems instead of rebuilding core infrastructure
The new architecture uses a Parent/Child model. The Global Agent (Parent) classifies the user's intent, then dynamically creates Child Agents as needed to distribute the work.
For example, when a complex request comes in to find contracts meeting specific conditions and assess their risk, the Global Agent establishes a plan and runs child agents in parallel that handle search, document extraction, and analysis. This dynamic agent creation approach enables flexible handling of complex tasks that are not predefined.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.