Highlights and Recording: Going further with the new Box Agent

Thanks to ​@rrajwade for yesterday’s overview and demos of Box Agent, also thanks to our guests who joined and shared in the Q&A. The session was an opportunity for users to ask questions and provide feedback on Box's AI journey.

Box Agent Overview

​@rrajwade provides a foundational overview of the Box Agent. The core value proposition is using AI with content to accelerate knowledge work. Key features discussed include:

• Recursive Box Agent: Instead of multiple single-purpose agents, Box has developed a single, powerful recursive agent capable of planning and iterating on complex, multi-step tasks. It consolidates capabilities like Q&A, search, and content generation.

• Box AI Studio: A platform for customers to build their own custom agents, providing them with specific knowledge, detailed instructions, and tools to perform specialized tasks.

• ProMode: An enhanced mode for any agent that enables a deeper level of thinking and synthesis for more complex work, going beyond simple Q&A.

• Intelligent Content Management Platform: All AI capabilities are built upon Box's core platform, ensuring that all operations adhere to existing security, compliance, data protection, and governance controls.

  

  

  

  

   

Main Topics: Use Case Demonstrations

The roundtable’s main segment features five detailed demonstrations of the Box Agent in action:

1. RFP Response Management: This demo showcases a custom agent that automates the creation of RFP responses. The process begins with an intake form, where an AI agent extracts key metadata from the uploaded RFP, assesses its fitness, and automatically creates a dedicated workspace. Another custom agent, trained on the company's historical wins and product documents, then generates a complete first draft of the response, which is then passed to a human for review and approval.

1. Contract Lifecycle Management (CLM): This walkthrough illustrates how AI agents streamline contract review. After a contract is generated from a template, a custom agent analyzes redlines from a third party. It references a playbook of past agreements to identify high-risk clauses, provides strategic recommendations to the legal reviewer, and can even draft revised language with correct legal formatting, significantly speeding up negotiations.

1. Insurance Claims Management: A vertical-specific solution where an AI agent assists claims adjusters. The demo shows an agent automatically extracting key information (like ICD-10 codes and provider details) from a medical document to populate fields in a claims system like Guidewire. The agent can also perform discrepancy analysis across multiple documents to flag potential fraud and answer complex legal questions by querying a curated knowledge hub of state insurance laws.

1. Lease Management: This use case focuses on synthesizing information from a complex series of documents. The agent reads an entire folder containing an original lease and all its subsequent amendments in chronological order. It reconstructs the 'current business truth' by identifying what has changed and what remains in effect, extracts the final terms as structured metadata, and updates a portfolio dashboard in real-time.

1. Document Libraries (Engineering Drawings): The final demo shows how Box Agent brings structure to a library of engineering drawings. An AI agent automatically reads the title block of a newly uploaded drawing to extract and apply metadata like drawing number and product line. It also demonstrates a powerful revision comparison feature, where the agent analyzes multiple versions of a drawing and generates a structured summary of the changes, including an impact and criticality analysis.

Q&A Session

The presenters answer several questions from the audience:

• On AI Document Validation: ​@rrajwade  clarifies that while AI can validate and flag documents, there is always a human-in-the-loop for final approval, ensuring the agent does not 'go rogue.' The agent's actions are governed by the permissions and context it is given.

• On the Importance of Metadata: Metadata is described as the 'secret sauce' that bridges unstructured content and structured intelligence. It creates a unified semantic layer for AI, powers precise search, automates downstream workflows, and is vital for scaling governance and security. It also helps ground AI models to reduce hallucinations.

• On API Access: API access for agents is not yet available but is on the product roadmap.

Resources

You can access the full recording here

What is AI metadata? Blog