An agent delegating a task to another agent

[arturgomesc]

I am creating a chatbot for salespeople who want to register a new task in an existing opportunity in my company's CRM. Creating an opportunity involves three steps, each of which makes an API call. To do this, I created an Agent for each of these steps, each of which uses the API call as a tool. Each Agent must be critical of the information the user provides, judging whether the response from the API call is valid and should move on to the next step.

However, I am having difficulty finding a way for one agent to delegate the next step to another. How can I do this? Any additional configuration, classifier prompt?

Here is some Python code that I used to try to do this using ChainAgent, but it always automatically passes the task from one Agent to another, without the first Agent defining whether to advance to the next Agent or not.

    get_opportunities_agent = BedrockLLMAgent(BedrockLLMAgentOptions(
        name="Get Opportunities Agent",
        streaming=False,
        description="Agent specialized in searching for existing opportunities in the company's CRM.",
        model_id="anthropic.claude-3-sonnet-20240229-v1:0",
        tool_config={
            'tool': get_opportunities_tool_description,
            'toolMaxRecursions': 5,
            'useToolHandler': get_opportunities_handler
        }
    ))

    get_opportunities_agent.set_system_prompt(get_opportunities_prompt)

    get_task_types_agent = BedrockLLMAgent(BedrockLLMAgentOptions(
        name="Get Task Types Agent",
        streaming=False, 
        description="Agent specialized in searching for task types. This Agent must be called after the Get Opportunity Agent.",
        model_id="anthropic.claude-3-sonnet-20240229-v1:0",
        tool_config={
            'tool': get_task_types_tool_description,
            'toolMaxRecursions': 5,
            'useToolHandler': get_task_types_handler
        }        
    ))

    task_types_prompt = get_task_types_prompt_with_opportunity()

    get_task_types_agent.set_system_prompt(task_types_prompt)

    create_task_options = ChainAgentOptions(
        name='Create Task Agent',
        description='Chain Agent responsible for creating tasks. It has two Agents: one that searches for opportunities and another that searches for task types.',
        agents=[get_opportunities_agent, get_task_types_agent],
        save_chat=True
    )

    chain_agent = ChainAgent(create_task_options)
    orchestrator.add_agent(chain_agent)