Building Your First Enterprise AI Agent with LangChain + Claude (Code Included)

Most “AI agent” tutorials show you a chatbot. This one shows you a production-ready, structured-output, audit-logged agent that actually meets enterprise expectations.

The Pattern: Tool-Use Agent with Bounded Outputs

The Agent (LangChain + Claude, Production-Ready)

from pydantic import BaseModel, Field
from langchain_anthropic import ChatAnthropic
from langchain.agents import create_tool_calling_agent, AgentExecutor
from langchain_core.prompts import ChatPromptTemplate
from langchain.tools import tool


class TicketTriageResult(BaseModel):
    """Output schema — the agent CANNOT respond outside this shape."""
    category: str = Field(description="One of: bug, feature, billing, security, other")
    priority: str = Field(description="One of: P0, P1, P2, P3")
    suggested_owner: str = Field(description="Team to route to")
    summary: str = Field(description="One-sentence summary")
    confidence: float = Field(ge=0, le=1)


@tool
def fetch_customer_tier(customer_id: str) -> str:
    """Look up the support tier for a customer."""
    # in production: query CRM
    return "enterprise"


@tool
def fetch_recent_tickets(customer_id: str) -> list[dict]:
    """Return the last 5 tickets for a customer."""
    return []  # in production: query ticket DB


def triage(ticket_text: str, customer_id: str) -> TicketTriageResult:
    llm = ChatAnthropic(model="claude-3-7-sonnet-20250219", temperature=0)
    structured = llm.with_structured_output(TicketTriageResult)

    prompt = ChatPromptTemplate.from_messages([
        ("system", "You are a support triage assistant. Use tools to gather context, then respond with the structured triage decision. Be conservative — escalate to P0 only when revenue or security impact is clear."),
        ("user", "Customer: {customer_id}nTicket:n{ticket_text}"),
        ("placeholder", "{agent_scratchpad}"),
    ])

    agent = create_tool_calling_agent(llm, [fetch_customer_tier, fetch_recent_tickets], prompt)
    executor = AgentExecutor(agent=agent, tools=[fetch_customer_tier, fetch_recent_tickets], verbose=False)
    raw = executor.invoke({"customer_id": customer_id, "ticket_text": ticket_text})

    # Validation gate
    result = structured.invoke(raw["output"])
    audit_log(ticket_text, customer_id, raw, result)
    return result