Google shows cross-language multi-agent workflow with ADK and A2A

Google has published a technical guide on building a crosslanguage multiagent pipeline using its Agent Development Kit (ADK) and the open Agent2Agent (A2A) protocol. The example combines a Python agent that extracts key terms from contracts with Gemini and a Go agent that performs deterministic compliance checks against corporate policy. The article explains why a single large agent can be difficult to maintain in production, citing issues such as context overload, wider failure impact, and limited testability. It presents a modular design in which each agent has a narrow role and communicates through shared protocol standards instead of languagespecific integrations. A2A is described as a discovery and messaging layer that lets agents built in different languages work together through Agent Cards and JSONRPC. Google’s RemoteA2aAgent abstraction in ADK can wrap a remote compliant service as a local subagent, allowing the Python orchestrator to coordinate extraction, compliance validation, and final reporting. The post also outlines an open source reference application and a demo interface that shows live handoffs between the two services, including a failure mode that routes work to manual review if the Go validator is unavailable.