Google explains why it built ADK 2.0

Google says ADK 2.0 was designed to help developers move AI agents from prototypes into production by combining agent flexibility with deterministic workflow control. The company says fully autonomous LLM agents can struggle with reliability in enterprise settings, including looping, skipping steps, or handling errors poorly. ADK 2.0 adds a workflow runtime that lets developers route tasks through code, mix tool calls with LLM steps, and keep execution paths more predictable. The post highlights a customer refund example to show how workflows can reduce token usage, lower latency, and limit context bloat compared with a fully autonomous agent loop. Google also says the design can help reduce promptinjection risk by separating language processing from execution control. ADK 2.0 is available for Python and Go, and Google positions it as a way to build hybrid agentic systems that use agents for reasoning while reserving workflows for predefined business logic.