Recording and Highlights - Learn about the new Box Agent

Thanks to Rutuja Rajwade and Matthew Terrell for yesterday’s overview of Box Agent. Also, thanks to our guests who joined and shared in the Q&A. Please see below for the recording and highlights.

Introduction to the new Box Agent
Rutuja Rajwade provides an overview of the new Box Agent. She explains that previous iterations involved several foundational agents (e.g., search, Q&A) for specific tasks. The new approach consolidates these into a single, singular Box Agent capable of planning, iterating, and executing complex requests based on user content.

Live Demos
Matthew Terrell leads a series of demos to showcase the agent's capabilities:
1. Basic Q&A: A simple query, "According to box company policy, do we have Memorial Day off?" is used to demonstrate how the agent securely scans the user's accessible content (like an employee handbook) to find and provide a grounded, accurate answer.

2. Analyzing with Box Hubs: The demo shows how Box Hubs, which are AI-enabled smart folders with pre-embedded content, allow the agent to rapidly parse and analyze thousands of files. A Hub is added as a knowledge source to the chat, and the agent is asked to summarize its contents, producing a detailed outline.

3. Custom Agents with AI Studio: This demo highlights how users can create specialized agents. A custom 'Marketing Agent' is configured with specific instructions and knowledge sources (e.g., brand guidelines). It is then used to analyze an uploaded document and generate a comprehensive report on its alignment with the company's marketing guidelines.

4. Content Creation (Beta): A preview of a beta feature using Claude skills from Anthropic shows the agent generating a new PowerPoint presentation. By providing a Hub of source content and a markdown file with presentation guidelines, the agent creates a complete, structured deck and saves it directly to Box.

Key Insights & Future Roadmap
Matthew Terrell discusses the underlying principles and future direction of the Box Agent.

• He emphasizes that the agent always operates within the invoking user's permissions, ensuring security and governance.
• A key part of the future roadmap is the development of 'Skills'—reusable, detailed scripts that the agent can run. This will enable more complex, one-off tasks and integrations, with plans to allow the agent to run code securely and utilize skills from other vendors.
• The integration of agents into Box Automate is highlighted with a loan processing workflow example, showing how agents can automate data extraction, validation, and routing, transforming manual processes into efficient, human-in-the-loop systems.

Q&A Session
The presenters answer audience questions on various topics:

• Metadata: The agent currently uses file activity and will soon incorporate extracted, formal metadata for more intelligent content curation in Hubs.
• Customization: Users can create custom agents with unique prompts and knowledge bases in AI Studio.
• Future of Agents: The goal is to unify capabilities into a single agent where users can toggle different skills and modes, rather than maintaining separate beta and GA versions.
• Costs: The Box Agent is included in enterprise plans with daily usage limits to ensure system health. An 'Expanded Mode' is available for pay-as-you-go usage beyond these limits, drawing from a pool of AI units.