LangGraph v1: 'create_react_agent' is deprecated — how do I move to langchain.agents.create_agent (prompt → system_prompt, AgentState, HumanInterrupt)?

Am I affected: you are if you import create_react_agent from langgraph.prebuilt, or use the prebuilt agent state and human-in-the-loop types, and have upgraded to LangGraph v1 / LangChain 1.x. create_react_agent is deprecated in favor of LangChain's create_agent, and several related classes moved or were removed. LangChain 1.x packages also require Python 3.10 or newer, so a pip install on 3.9 resolves to old versions or fails. How to check: run your app with warnings visible (python -W default) and look for deprecation warnings from langgraph.prebuilt, then grep for create_react_agent, AgentStatePydantic, HumanInterrupt, ValidationNode and MessageGraph. Fix (per LangChain's v1 migration guide): - from langgraph.prebuilt import create_react_agent → from langchain.agents import create_agent - create_react_agent(model, tools, prompt=...) → create_agent(model, tools, system_prompt=...) - AgentState, AgentStatePydantic, AgentStateWithStructuredResponse(Pydantic) → langchain.agents.AgentState - HumanInterruptConfig / ActionRequest → langchain.agents.middleware.human_in_the_loop.InterruptOnConfig; HumanInterrupt → HITLRequest from the same module - ValidationNode: removed; create_agent validates tool calls itself - MessageGraph: removed; use StateGraph with a 'messages' key - Python: upgrade to 3.10+ Hooks you previously added with pre/post-model nodes now belong in create_agent's middleware. Interim: pin langgraph below 1.0 (and the matching langchain-core) until you can migrate. Pin explicitly, because loose ranges pull in v1. After migrating: run the agent's tests end to end, especially tool-calling loops, structured output and interrupts/resume. Check that checkpoints saved before the upgrade still load with your checkpointer. Sources: https://docs.langchain.com/oss/python/migrate/langgraph-v1 https://docs.langchain.com/oss/python/release-policy