How Box built its AI agent with LangGraph
Blog post from Box
Box transitioned its AI agent platform to a multi-step agentic system using LangGraph as its core execution engine, opting against building a custom solution from scratch. This decision enabled the development of the Box Agent platform, which is structured into five distinct layers: entry points, an intelligence service, an Agent Orchestrator, an LLM gateway, and model providers. LangGraph's graph-based execution model supports branching, parallel execution, and real-time event streaming, allowing for dynamic task planning and seamless execution at an enterprise scale. Agents, defined using Box’s Agent Definition Language, are compiled into Deep Agents at runtime, facilitating behavior updates without code deployments. The platform's architecture supports resumability, enabling interrupted sessions to be resumed from the latest checkpoint. The system's modularity and scalability allow it to handle both simple and complex workflows, with the Agent Orchestrator managing the full lifecycle of agent execution across multiple teams at Box. The integration with LangGraph and the use of Deep Agents provide a robust framework for handling dynamic tasks and live configuration changes, ensuring reliability and developer velocity without sacrificing the complexity of managing multiple provider integrations.
No tracked trend matches for this post yet.
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.