Skip to main content

Lab 16 Solution: Building a Hybrid News Aggregator

Goal​

This file contains the complete code for the agent.py script using the ADK 2.0 Hybrid Graph pattern (Sequential + Parallel).

news_aggregator/agent.py​

from google.adk import Agent, Workflow
from google.adk.workflow import JoinNode

# 1. Define Specialist Nodes
tech_researcher = Agent(
    name="tech_researcher",
    model="gemini-3.5-flash",
    instruction="Find 3 exciting headlines about AI and Robotics. Be concise.",
    output_key="tech_news"
)

market_researcher = Agent(
    name="market_researcher",
    model="gemini-3.5-flash",
    instruction="Find 3 key headlines about Stock Market trends. Be concise.",
    output_key="market_news"
)

summarizer = Agent(
    name="summarizer",
    model="gemini-3.5-flash",
    instruction="""
    You are a news editor. Create a brief newsletter using the data provided:
    TECH: {tech_news}
    MARKET: {market_news}
    
    Synthesize the information into a single, cohesive daily briefing.
    """
)

# 2. Define the Synchronization Point
syncer = JoinNode(name="news_sync")

# 3. Assemble the Workflow
# Parallel Fan-out + Sequential Fan-in
root_agent = Workflow(
    name="NewsSystem",
    edges=[
        # Both start at the same time and connect to the same JoinNode
        ("START", tech_researcher, syncer),
        ("START", market_researcher, syncer),
        
        # Once both are done, the syncer triggers the summarizer
        (syncer, summarizer)
    ]
)

Self-Reflection Answers​

  1. What happens if one of the parallel researchers fails?

    • Answer: By default, the JoinNode waits for ALL incoming edges. If one fails, the JoinNode will not fire its outgoing edge, potentially stalling the workflow. In production, you would add retry logic or error handling to ensure the graph can still converge.

  2. Can I add a third researcher (e.g., 'sports_researcher')?

    • Answer: Yes! You just define the agent and add a third edge: ("START", sports_researcher, syncer). The syncer will automatically wait for all three.

  3. Why use output_key if data flows automatically?

    • Answer: While ADK 2.0 passes the output of one node to the next, a JoinNode receives multiple outputs. Using output_key saves individual results into the global session state, making it much easier for the summarizer to access specific variables like {tech_news} via string interpolation.