create_agent: LangChain's prebuilt agent in one call

The langchain-core post showed the agent loop written by hand. create_agent is the layer that writes it for you: give it a model, tools, and a system prompt, and the loop arrives prebuilt in one function call — much like the OpenAI Agents SDK. It's the fast path from the ecosystem-layers ladder.

One call, the loop prebuilt

from langchain.agents import create_agent

agent = create_agent(
    model='openai:gpt-4o',
    tools=[get_weather],
    system_prompt='You are a concise, helpful assistant.',
)
result = agent.invoke({'messages': [{'role': 'user', 'content': 'weather in Paris?'}]})
# await agent.ainvoke(...) for async — nothing else changes

It's a LangGraph graph underneath

The important detail: create_agent returns the same compiled LangGraph graph you'd have wired by hand a layer down. So you inherit LangGraph's durability — checkpointing, resumable runs, streaming — for free, and you can drop into the graph when you need more control. You climbed a layer; you didn't leave the ecosystem.

Batteries included

• Tools — your @tool functions from langchain-core; the loop runs itself.
• Memory — a checkpointer and a thread_id, exactly like LangGraph, so conversations persist and resume.
• Structured output — response_format returns your Pydantic object instead of text.
• Middleware — your own hooks inside the loop, to observe or modify each step.
• MCP — connect MCP server tools so the agent can drive real systems (see the MCP posts).

When to use it

Reach for create_agent when you want a production-ready agent quickly without wiring the loop yourself — the same trade as the OpenAI Agents SDK's one-call agent. Drop down to langgraph or langchain-core when you need bespoke control over the flow, and up to deep agents when the task needs planning and sub-agents.

create_agent is the loop you'd write by hand, minus the writing — and because it's a LangGraph graph underneath, dropping to a lower layer is always one step away.