Thanks to William Higgins,
Summary of the main sections
Section 1 - AI Studio Enhancements
Recording clip here
Section 1 - AI Agent Demo #1 - Legal Contract Analysis
-
Risk Detection: Automatically flags nonstandard clauses and missing protections.
-
Intelligent Routing: Identifies stakeholders who need to review specific items.
-
Speed: Completes hours of manual work in seconds.
-
Executive Summaries: Offers quick, detailed overviews of contract content.
The live demo illustrated the agent in action, analyzing clauses for risk levels and providing actionable recommendations—such as flagging vendor-favoring liability clauses—to streamline the review process.
Section 1- AI Agent Demo #2 - Fraud Detection and Financial Services Optimization
Demo #2 showcased an AI KYC Agent built to accelerate document analysis in financial services. The agent scans patterns in bank statements—such as large transactions or transfers to unknown entities—and categorizes them by risk level. Key benefits include:
-
Prioritized Findings: Results are sorted by risk severity for efficient review.
-
Rapid Analysis: Complex reviews completed in seconds, compared to hours with manual methods.
This augmentative AI approach combines automation with human expertise for high-quality outcomes in underwriting and fraud detection.
Section 2- Hands-On Agent Building: Creating Custom Box AI Agents
Recording clip here
The session transitioned into building custom AI agents within Box AI Studio. The step-by-step walkthrough offered participants practical experience with two foundational exercises: creating a "Quiz me" agent or a "risk flagging" agent.
Participants explored the interface and learned how to configure prompts, navigate the admin console, and test features using the playground. “Quiz me,” an agent designed to generate interactive quizzes, was highlighted for its practicality and fun.
Testing and Iteration
William demonstrated how to refine and test AI models by switching from the default configuration to Claude Sonnet 4.5:
“Let’s see how Claude SONNET 4.5 does. It seems to work just as well or better, maybe.”
This iterative process in the playground enabled users to evaluate agent responses and optimize their functionality before deployment.
Section 3 - Upcoming Enhancements and Roadmap Highlights
Recording clip here
Box AI Agents Across All Content
“Part of our new Box AI experience that's coming in Q4, where you'll be able to leverage agents across all of your Box content.”
Advanced Search and Research Agents
Box AI’s search agent tools were explained in detail, emphasizing their ability to dive deep into content for research-focused summarization and data extraction. Ideal for compiling reports or analyzing clinical trials, these agents bring efficiency and depth to content-heavy workflows.
Checkout the Box AI Agent template library at the bottom of this page here
Unique Advantages of Box AI
Model Agnostic
As highlighted by
“One of the really distinct benefits of Box AI is the fact that we are model agnostic… having access to all those models is really beneficial.”
This flexibility ensures users can pick the most suitable model for their needs, whether for document review, fraud detection, or other applications.
External API Integration
Box AI is ready for external integrations. Agents can be deployed via APIs and are compatible with MCP (Model Context Protocol), enhancing their adaptability across diverse workflows.
Next Generation of Metadata Extraction
A new Box AI UI—Box Extract—is in the pipeline. This feature will allow users to define custom metadata templates and use AI prompts for extraction and classification. As William noted: “For automatically classifying or tagging documents…we would recommend Box AI extract agents.”
Ready for more?
Join our upcoming roundtable: