Google ADK Go 2.0 points agent builders toward explicit graphs

Google's ADK Go 2.0 announcement emphasizes graph-based multi-agent workflows, human-in-the-loop checkpoints, retries, dynamic Go execution, and a shared runtime for simple and graph agents. Why it matters: Production agent systems need control flow, recovery, and review boundaries. A graph-first model makes those seams visible instead of hiding them inside a single prompt loop.

Try this: Use the post as an architecture checklist: draw the graph state, failure paths, human checkpoints, retry policy, persistence, and telemetry before adding another agent step.