How to build an agent with LangChain

Define one typed tool

A good first agent has a narrow job. Make tool inputs explicit and keep credentials inside the tool implementation rather than putting them in prompts.

from langchain.tools import tool

@tool
def lookup_invoice(invoice_id: str) -> str:
    """Return the status and balance for an invoice."""
    return billing_api.get_invoice(invoice_id)

Create and invoke the agent

Bind the tool to a supported chat model and give the agent a short system instruction that describes success and important boundaries.

from langchain.agents import create_agent

agent = create_agent(
    model="openai:gpt-5",
    tools=[lookup_invoice],
    system_prompt="Help finance staff inspect invoices. Never modify billing data.",
)

result = agent.invoke({
    "messages": [{"role": "user", "content": "Check invoice inv_1042"}]
})

Make it production-ready

Add tracing, retries around transient failures, a maximum step count, and authorization outside the model. Test normal requests as well as prompt injection and malformed tool arguments.