How Box AI Went Agent-Native With LangChain Deep Agents

Source: LangChain — 2026-07-08

Summary

LangChain details how Box built its enterprise content-agent platform on LangChain's Deep Agents framework, using a parent/child "Global Agent" architecture, model-agnostic routing, prompt caching, and explicit context-window management to handle enterprise document search and synthesis at scale.

Key Takeaways

• Uses a parent/child "Global Agent" architecture rather than a single flat agent loop, to manage complex enterprise content workflows.
• Model-agnostic routing lets Box swap underlying models without re-architecting the agent stack.
• Explicit context-window management and prompt caching called out as necessary, not optional, at enterprise document-scale.
• A concrete production case study of Deep Agents beyond LangChain's own launch/marketing material.

Discussion

(No questions yet — ask follow-ups via a Claude Code chat session on this repo; answers get appended here.)