Building agents with the ADK and the new Interactions API

This post explores two integration patterns for using Google's new Interactions API alongside the Agent Development Kit (ADK) and Agent2Agent (A2A) protocol.

• •The Interactions API provides a unified endpoint for both raw Gemini models and built-in agents like the Gemini Deep Research Agent.
• •Pattern 1 uses the Interactions API as the inference engine for ADK agents, enabling server-side state management via previous_interaction_id and background execution mode.
• •Pattern 2 uses InteractionsApiTransport to treat an Interactions API endpoint as a standard remote A2A agent, requiring no changes to existing A2A client code.
• •The bridge maps A2A protocol operations (SendMessage, Task, TaskStatus) directly to Interactions API equivalents.
• •Configuration smuggling via A2A extensions allows passing specific settings like thinking_summaries inside AgentCard without breaking the standard protocol.