Box explains its Deep Agent architecture: parent-child agents over a LangGraph orchestrator
LangChain · x · 2026-10-28
Box's Deep Agent architecture uses a parent/child model: simple requests are handled directly by the parent with no planning, while complex requests spawn child agents to distribute work — all behind a uniform invocation surface. The diagram shows a LangGraph-based Agent Orchestrator with a Global Agent and SubAgents, plus Tools, Middleware and Guardrails, routing through an LLM Gateway to multiple providers, with AI Center AX, AI APIs and Box MCP as entry points.
Related event: Box Reveals Deep Agent Architecture with Dynamic Parent-Child Agents(2 posts)→
More from coding & agent
- Judgment Labs launches publicly: monitoring and eval layer for long-horizon agents — testingcatalog · 2026-10-28
- OSU NLP Group Wins Three Paper Awards Same Day at EMNLP 2026 and NeurIPS Workshop — hhsun1 · 2026-10-28
- Replit launches Cloud Agents: non-engineers ship reviewed PRs from natural language — Jeff_Burke14 · 2026-10-28
- Sequoia and OpenAI host $100K poker bot tournament where agents build bots in 60 minutes — polynoamial · 2026-10-28
- PPCC26 demos praised: real orgs like NFL, real code live in VS Code, no fake companies — donasarkar · 2026-10-28
- Five principles for curating LLM agent evals: diverse, realistic, hard, discriminative, high-quality — cwolferesearch · 2026-10-28