Box (BOX) Investor update summary

AI strategy and market

• AI agents are positioned to expand from assistants into autonomous fleets that execute broad, content-heavy workflows across industries.

• Unstructured content is described as 90% of enterprise data, including contracts, product plans, financial projections and employee information.

• More than 60% of Fortune 500 organizations are customers; the strategy centers on securely connecting enterprise content and context to AI agents.

Context retrieval and models

• Box Context Retrieval adds managed indexing and retrieval for enterprise content, automatically updating indexes as files change and enforcing existing Box permissions.

• Retrieval combines lexical and semantic search with rank fusion, surfacing relevant information within files for Box Agent and third-party agents through MCP.

• Recall benchmark accuracy improved by over 20%; retrieving relevant document sections reduces token use by ~90% and enables answers 4x faster.

• NVIDIA Nemotron retrieval models support Context Retrieval, while NVIDIA open-weight Nemotron models expand model choices for Box Agents.

Agentic workflows and automation

• Box Agent supports complex unstructured-content tasks; planned additions include safe web access, reusable skills, work lasting up to 20 hours and file-based deliverables.

• Planned Box Extract enhancements include evaluation sets, human review for potential issues and adaptive model routing; Box Automate orchestrates people, agents and enterprise systems with guardrails.

• Box Architect is an autonomous workflow builder that proposes designs from business goals, creates agents, metadata templates, apps and workflows, and is intended to optimize processes over time.

• Box Architect is positioned as an outer layer above workflow tools; early customer interest points to automating a broader range of processes as implementation becomes easier.