Google ADK 2.0 tutorial: workflows vs ADK 1.0

Google ADK 2.0 is the focus of this AI agent tutorial, which compares its workflow controls with ADK 1.0 through three Python examples. The presenter explains graph routing, coordinator handoffs and workflows that pause for human input.

The support router uses an LLM to classify messages as bugs, billing questions or feature requests. Code defines the edges that route each result to a handler. The presenter contrasts this with a prompt-driven ADK 1.0 approach and argues that moving routing into code makes execution more predictable. The example still relies on the model to classify the message correctly.

A travel planner demonstrates two subagent behaviors. The weather checker runs once and returns control to the coordinator. The flight booker can ask follow-up questions before completing its task and handing control back. The demo includes a request that uses both subagents.

The refund example uses dynamic workflows with ordinary Python conditions. The presenter explains a node decorator, input requests and resume data, alongside automatic checkpointing. A $50 refund receives automatic approval; a $350 refund pauses for human approval before processing continues.

The examples run through ADK's web development interface. The supplied material does not establish local model support, offline operation or hardware requirements.