langchain-ask-human example is using deprecated design

[jleonelion]

This example calls initialize_agent function: https://github.com/Chainlit/cookbook/blob/main/langchain-ask-human/app.py#L55
But that is deprecated: https://github.com/langchain-ai/langchain/blob/2ab5673eb1eae4a8945ed090d81b6f47e4639d58/libs/langchain/langchain/agents/initialize.py#L15-L22
In fact, the alternative suggested create_react_agent is also not recommended: https://github.com/langchain-ai/langchain/blob/2ab5673eb1eae4a8945ed090d81b6f47e4639d58/libs/langchain/langchain/agents/react/agent.py#L30-L34
And using langgraph prebuilt is recommended: https://github.com/langchain-ai/langgraph/blob/6906e12edb551162f5fede5d9f55b4bb206bb59c/libs/langgraph/langgraph/prebuilt/chat_agent_executor.py#L197

All in all, this example is inspirational, but doesn't really hit the mark in terms of a recommended approach for building a ReAct agent with LangChain/LangGraph

[jleonelion]

I've made good progress on implementing the LangGraph aspects using pattern here: https://github.com/langchain-ai/langchain-academy/blob/main/module-1/agent.ipynb

And I added the HumanInputChainlit tool (with some logging)

App logs show the tool class being called but I am never prompted in the UI nor do I see log message to indicate response is returned...it's just hanging.

   logger.debug(f"query: {query}")
    res = run_sync(cl.AskUserMessage(content=query).send())
    logger.debug(f"response: {res}")

I have verified the LangGraph logic by creating a Jupyter notebook (that approach uses the built-in input instead of AskUserMessage) - this problem is unique to running in the Chainlit server.