Google ADK tutorial: build a blog writing agent

Smitha Kolan explains agent workflows and walks through a Python blog writing example using Google's Agent Development Kit (ADK). She introduces ReAct as a cycle of reasoning, action and observation, then compares sequential, reactive and planning patterns by the kinds of tasks they suit.

The setup starts with installing uv and Google ADK, then creating agent.py. A planner uses a model selected through an environment file to produce a Markdown outline with a title, introduction, four to six sections and a conclusion. It saves the outline in shared state for the writer to use.

Separate validation agents check the outline and finished post. Kolan describes loops that rerun each generation step when its checker requests a retry, with a limit of three attempts. These checks concern structure and requested explanations; the tutorial does not establish factual accuracy of the generated article.

A root AI agent calls the planner and writer through two exposed tools, then adds three alternative titles and two short hooks. The walkthrough ends with launching the ADK web UI and submitting a topic through a local browser address. That local interface does not establish that model inference runs on the user's hardware. The transcript does not specify a model version or demonstrate offline operation.