Build an ADK Agent

• Key points are in the visual diagram above.

Google's kit is how Gemini-native teams ship agents without re-inventing orchestration on Vertex.

A Gemini specialist with a toolkit and junior agents they can call, scheduled on Google's runtime.

Scenario

Coordinator agent with search tool and a 'refund' sub-agent. Runner.run(user_id, session_id, message).

Solution

In Google ADK, apply Build an ADK Agent to this scenario: Coordinator agent with search tool and a 'refund' sub-agent. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Build an ADK Agent (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Build an ADK Agent happens in the code.

1from google.adk.agents import LlmAgent  # import dependencies2from google.adk.runners import Runner  # import dependencies3
4refund = LlmAgent(name="refund", model="gemini-2.5-flash", instruction="Handle refunds only.")  # key line for Build an ADK Agent5root = LlmAgent(  # key line for Build an ADK Agent6    name="desk",7    model="gemini-2.5-flash",8    instruction="Route billing to refund. Else answer.",9    sub_agents=[refund],  # key line for Build an ADK Agent10    tools=[search_tool],11)12runner = Runner(agent=root)  # key line for Build an ADK Agent13# runner.run(user_id="u1", session_id="s1", new_message="refund order 99")

• Copying a LangGraph into ADK 1:1
• No session id

Quick recap — the most important points from this module.