Chapter 8 — Multi-Agent Systems

# Supervisor pattern in LangGraph
from langgraph.graph import StateGraph, START, END

def supervisor(state):
    """Decides which agent to route to next."""
    response = llm.invoke(f"""
    You are a supervisor managing these agents: researcher, coder, writer.
    
    Current task: {state["task"]}
    Progress so far: {state["results"]}
    
    Which agent should work next? Or should we finish?
    Respond with: researcher, coder, writer, or FINISH
    """)
    return {"next_agent": response.content.strip().lower()}

def researcher(state):
    """Research agent with web search tools."""
    result = research_agent.invoke(state["task"])
    return {"results": state["results"] + [result]}

# Build the graph
graph = StateGraph(AgentState)
graph.add_node("supervisor", supervisor)
graph.add_node("researcher", researcher)
graph.add_node("coder", coder)
graph.add_node("writer", writer)

graph.add_edge(START, "supervisor")
graph.add_conditional_edges("supervisor", route_to_agent)

# Swarm pattern (OpenAI Agents SDK style)
from agents import Agent, handoff

triage_agent = Agent(
    name="Triage",
    instructions="Route the customer to the right department.",
    handoffs=[
        handoff(tech_support_agent, "Technical issues"),
        handoff(billing_agent, "Billing questions"),
        handoff(returns_agent, "Returns and refunds"),
    ]
)

tech_support_agent = Agent(
    name="Tech Support",
    instructions="Help users with technical problems.",
    tools=[search_docs, check_status],
    handoffs=[handoff(triage_agent, "Not a tech issue")],
)