3 - 6 Hours 3 Modules
Main Topic: AI for Developers
Workload: 3 - 6 Hours
Level: Advanced
Target Audience: Software Architects, AI Engineers and Innovation Leaders
Conceptualization of the hierarchical model. Learning how to define the "Manager" agent responsible for breaking down the user's request and delegating the parts to the specialized "Worker" agents, ensuring that each one focuses only on its area of expertise.
The science of baton passing. Implementation of communication protocols (such as the Model Context Protocol - MCP) to ensure that the output of one agent (JSON) is the perfect input for the next, eliminating context loss and integration errors between system layers.
Long-lasting systems laboratory. How to set up agents that run in the background for hours, performing deep searches and automatic reviews. Discussion about system modularity: how to replace or update a single agent (e.g., swapping the Researcher) without compromising the entire structure.
from langchain.tools import tool
from langchain.agents import create_agent
# Criar um sub-agente
subagent = create_agent(model="anthropic:claude-sonnet-4-20250514", tools=[...])
# Transformando um sub-agente em ferramenta
@tool("research", description="Research a topic and return findings")
def call_research_agent(query: str):
result = subagent.invoke({"messages": [{"role": "user", "content": query}]})
return result["messages"][-1].content
# Agente principal com o sub-agente como uma ferramenta
main_agent = create_agent(model="anthropic:claude-sonnet-4-20250514", tools=[call_research_agent])