How to prevent duplicate AI responses when using Command.goto for multi-agent handoffs?

OSS Product Help LangChain
1

Hey folks! I’m building a multi-agent system with LangGraph and running into duplicate responses when delegating between agents.

Architecture:

  • Main graph with 2 agents (orchestrator + specialist)

  • Both created with langchain.agents.create_agent

  • Orchestrator handles queries OR delegates to specialist

  • Graph: START → orchestrator → [END or specialist] → END

The Problem:
When user asks for something the specialist handles:

  1. Orchestrator calls handoff tool

  2. Orchestrator generates response: “I’ll help you with that…”

  3. Specialist ALSO generates response: “I’ll help you with that…”

  4. User sees BOTH messages (duplicate)

What I’ve Tried:

Attempt 1 - Conditional edges:

  • Tool sets active_workflow state

  • Conditional edge routes based on state

  • Result: Orchestrator still generates final response before routing

Attempt 2 - Command.goto:

@tool
def handoff_to_specialist(runtime: ToolRuntime) -> Command:
    return Command(
        goto="specialist",
        graph=Command.PARENT,
        update={"messages": state["messages"] + [transfer_message]}
    )
  • Result: Still seeing orchestrator’s response + specialist’s response

Question:
With create_agent, is there a way to suppress the orchestrator’s final response when Command.goto is used? Or is this a fundamental limitation where agents always generate responses after tool calls?

The goal: Only specialist’s response should reach the user, not orchestrator’s.

Any guidance appreciated!