framework-selection

Answer the decision questions in order: use Deep Agents if you need sub-task planning, file management, or persistent memory; LangGraph if you need custom control flow like loops and branching; LangChain for single-purpose agents or simple chains and retrieval pipelines.

Use LangChain for focused, self-contained tasks like RAG pipelines, model calls, and single-purpose tool agents. Use LangGraph when you need to own the control flow with loops, dynamic branching, parallel workers, or human-in-the-loop steps.

Yes, a LangGraph compiled graph can be registered as a subagent inside Deep Agents. The Deep Agents orchestrator delegates to it via the task tool without needing to know its internal graph structure.

Deep Agents ships with TodoListMiddleware, FilesystemMiddleware, and SubAgentMiddleware always enabled. SkillsMiddleware, MemoryMiddleware, and HumanInTheLoopMiddleware are opt-in additions you configure as needed.

No, LangGraph has no middleware layer. You wire behavior directly into nodes and edges, while Deep Agents provides an explicit configurable middleware layer and LangChain only offers callbacks.

Avoid Deep Agents when the task is simple enough for a single-purpose LangChain agent, or when you need precise hand-crafted control over every graph edge, in which case LangGraph directly is the better fit.