Highlights and Recording: Going further with the new Box Agent

Thanks to ​@Meena Ganesh Box and Roger Schmitt for yesterday’s overview and demos of Box Agent and AI Home.

Full recording

 scroll down for recordings of individual demos

Overview of the Box Agent

​@Meena Ganesh Box provided a detailed overview of the Box Agent, which launched in the summer. Key points included:

• Purpose: To accelerate knowledge work by unlocking the 90% of enterprise content that is unstructured. The goal is to move beyond simple keyword search to 'searching with intent,' allowing AI to understand the user's ultimate goal and synthesize information accordingly.
• Capabilities: The Box Agent is a single, multi-purpose agent that can analyze prompts, plan and execute a series of tasks, and leverage Box's workflow tools. It can create and compose drafts, briefs, and emails grounded in the user's own secure files.
• Integration: The agent is accessible across various Box surfaces, including Box Hubs, Automate, and AI Home, with a persistent session history that carries context between them.
• Customization with Box AI Studio: ​@Meena Ganesh Box  introduced Box AI Studio, the agent builder solution that allows organizations to create custom agents tailored to specific workflows. She provided an example of a customer who built an RFP workflow agent with specific instructions on which knowledge sources to use, response formatting, and guardrails to prevent hallucinations.
• Security and Governance: A core theme was that the Box AI platform is built on Box's trusted content layer, inheriting its robust permission structure. The platform's 'secure RAG' (Retrieval-Augmented Generation) approach performs a permissions check as the very first step, ensuring users can only access information they are authorized to see.

Demos

Roger Schmitt presented a series of in-depth demos showcasing the Box Agent's capabilities in various business scenarios:

RFP Response Management: This demo illustrated a fully automated workflow. An incoming RFP was ingested via a form, a custom AI agent extracted key metadata to assess its 'fitness,' and the document was automatically routed. Another custom agent, trained on the company's knowledge base, then generated a complete draft response, which was sent for human review and approval before being tracked on a metadata-driven dashboard.

Question on RFP: When using Box AI Studio to build a dedicated 'RFP Agent', what are the best practices for structuring Hubs or document repositories so the agent prioritizes official, approved answers over draft proposals?

Answer: Using a Box Hub for final version of the RFP only as a curated content portal of RFPs  and point the users to only use Box AI to that Hub  🎉 

Contract Lifecycle Management (CLM): The demo showed how a new contract could be generated from a simple form. When the client returned a redlined version, a custom AI agent analyzed the changes, identified high-risk clauses by comparing them to past 'gold standard' agreements, and escalated the contract to legal. The agent then provided strategic recommendations and drafted revised legal language, dramatically speeding up the negotiation cycle before the final version was sent for signature via Box Sign.

Question on CLM : How are teams using Box for CLM day to day?

Answer: Day to day, teams are using Box AI to do the heavy lifting on contracts. Think pulling out tricky renewal dates or key terms without reading 50 pages, spotting redlines instantly, and having an AI agent answer questions across all active agreements. A great example is State Street Bank - they used Box to automate client agreements and transaction confirms, scaling up to 8,000+ accounts while saving a ton of time and cash

Claims Management: Integrating with Guidewire Claims Center, this demo showed a medical document being uploaded. The Box Agent automatically extracted key data (like ICD-10 codes), populated the fields in Guidewire, and applied security classifications and retention policies. A custom 'discrepancy analysis' agent was then used to cross-reference multiple documents to identify potential fraud indicators.

Question: Is the Guidewire and Box integration similar to the Salesforce and Box integration? Where metadata can automatically be extracted using AI from documents and a new Salesforce records can be created or a field can be updated?

Also for SAP and Box Integration via VersaFile - where metadata can be extracted from documents and write back to SAP record / fields 

Answer: Yes these are similar: https://support.box.com/hc/en-us/articles/46329311590291-Using-the-Box-AI-Data-Extraction-for-Salesforce-Integration

Lease Management: This powerful demo showed a 'lease abstraction' agent analyzing a folder containing an original lease and its seven subsequent amendments. The agent synthesized all the documents to create a single, up-to-date abstract, extracted the critical data as metadata, and automatically updated a portfolio dashboard. This enabled users to ask complex questions in natural language and receive instant, cited answers.

Document Libraries (Engineering Drawings): The final demo focused on bringing structure to a high volume of technical documents. Upon uploading a new drawing revision, AI extracted metadata from the title block, automated file routing, and managed version control. A custom agent was then used to compare all three revisions of a drawing, generating a structured summary of changes, an impact analysis, and a criticality rating.

Q&A and Conclusion

The session included questions about best practices for structuring knowledge sources for agents and the capabilities of various integrations (Salesforce, SAP).

A final poll on desired future improvements showed an even split between allowing more users to build agents, adding web search, and enabling agents to produce finished documents. 

Questions or feedback on Box Agent and AI home? Please share in the replies.